| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}})] | |
| { | |
| func main<ios16>(tensor<fp32, [1, 77]> input_ids) { | |
| tensor<int32, []> var_5 = const()[name = tensor<string, []>("op_5"), val = tensor<int32, []>(-1)]; | |
| tensor<bool, []> var_6 = const()[name = tensor<string, []>("op_6"), val = tensor<bool, []>(false)]; | |
| tensor<string, []> cast_1_dtype_0 = const()[name = tensor<string, []>("cast_1_dtype_0"), val = tensor<string, []>("int32")]; | |
| tensor<int32, []> inputs_embeds_axis_0 = const()[name = tensor<string, []>("inputs_embeds_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> inputs_embeds_batch_dims_0 = const()[name = tensor<string, []>("inputs_embeds_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [49408, 768]> text_encoder_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor<fp16, [49408, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<int32, [1, 77]> cast_2 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor<string, []>("cast_2")]; | |
| tensor<fp16, [1, 77, 768]> inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_2, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor<string, []>("inputs_embeds_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> position_embeddings_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [44352]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75890816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75935232))), name = tensor<string, []>("position_embeddings_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 768]> input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16_palettized)[name = tensor<string, []>("input_3_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75935424)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75937024)))]; | |
| tensor<fp16, []> var_15_to_fp16 = const()[name = tensor<string, []>("op_15_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75938624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76381056))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76381248)))]; | |
| tensor<fp16, [1, 77, 768]> linear_0_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; | |
| tensor<fp16, []> var_106_to_fp16 = const()[name = tensor<string, []>("op_106_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_5_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_106_to_fp16)[name = tensor<string, []>("tensor_5_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76382848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76825280))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76825472)))]; | |
| tensor<fp16, [1, 77, 768]> linear_1_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")]; | |
| tensor<int32, [4]> var_111 = const()[name = tensor<string, []>("op_111"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_112_cast_fp16 = reshape(shape = var_111, x = linear_1_cast_fp16)[name = tensor<string, []>("op_112_cast_fp16")]; | |
| tensor<int32, [4]> var_113_perm_0 = const()[name = tensor<string, []>("op_113_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76827072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77269504))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77269696)))]; | |
| tensor<fp16, [1, 77, 768]> linear_2_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")]; | |
| tensor<int32, [4]> var_118 = const()[name = tensor<string, []>("op_118"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_119_cast_fp16 = reshape(shape = var_118, x = linear_2_cast_fp16)[name = tensor<string, []>("op_119_cast_fp16")]; | |
| tensor<int32, [4]> var_120_perm_0 = const()[name = tensor<string, []>("op_120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_127 = const()[name = tensor<string, []>("op_127"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_128_cast_fp16 = reshape(shape = var_127, x = tensor_5_cast_fp16)[name = tensor<string, []>("op_128_cast_fp16")]; | |
| tensor<int32, [4]> var_129_perm_0 = const()[name = tensor<string, []>("op_129_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_131 = const()[name = tensor<string, []>("op_131"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_47 = transpose(perm = var_129_perm_0, x = var_128_cast_fp16)[name = tensor<string, []>("transpose_47")]; | |
| tensor<fp16, [12, 77, 64]> query_states_1_cast_fp16 = reshape(shape = var_131, x = transpose_47)[name = tensor<string, []>("query_states_1_cast_fp16")]; | |
| tensor<int32, [3]> var_133 = const()[name = tensor<string, []>("op_133"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_46 = transpose(perm = var_113_perm_0, x = var_112_cast_fp16)[name = tensor<string, []>("transpose_46")]; | |
| tensor<fp16, [12, 77, 64]> key_states_3_cast_fp16 = reshape(shape = var_133, x = transpose_46)[name = tensor<string, []>("key_states_3_cast_fp16")]; | |
| tensor<int32, [3]> var_135 = const()[name = tensor<string, []>("op_135"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_45 = transpose(perm = var_120_perm_0, x = var_119_cast_fp16)[name = tensor<string, []>("transpose_45")]; | |
| tensor<fp16, [12, 77, 64]> value_states_3_cast_fp16 = reshape(shape = var_135, x = transpose_45)[name = tensor<string, []>("value_states_3_cast_fp16")]; | |
| tensor<bool, []> attn_weights_1_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_1_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_1_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_1_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_1_cast_fp16 = matmul(transpose_x = attn_weights_1_transpose_x_1, transpose_y = attn_weights_1_transpose_y_1, x = query_states_1_cast_fp16, y = key_states_3_cast_fp16)[name = tensor<string, []>("attn_weights_1_cast_fp16")]; | |
| tensor<int32, [4]> var_140 = const()[name = tensor<string, []>("op_140"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_141_cast_fp16 = reshape(shape = var_140, x = attn_weights_1_cast_fp16)[name = tensor<string, []>("op_141_cast_fp16")]; | |
| tensor<fp16, [1, 1, 77, 77]> op_56_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4447]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77271296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77275840))), name = tensor<string, []>("op_56_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_3_cast_fp16 = add(x = var_141_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_3_cast_fp16")]; | |
| tensor<int32, [3]> var_146 = const()[name = tensor<string, []>("op_146"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_5_cast_fp16 = reshape(shape = var_146, x = attn_weights_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_7_cast_fp16 = softmax(axis = var_5, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; | |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7_cast_fp16, y = value_states_3_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")]; | |
| tensor<int32, [4]> var_151 = const()[name = tensor<string, []>("op_151"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_3_cast_fp16 = reshape(shape = var_151, x = attn_output_1_cast_fp16)[name = tensor<string, []>("attn_output_3_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_5_perm_0 = const()[name = tensor<string, []>("attn_output_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_154 = const()[name = tensor<string, []>("op_154"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_44 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast_fp16)[name = tensor<string, []>("transpose_44")]; | |
| tensor<fp16, [1, 77, 768]> input_9_cast_fp16 = reshape(shape = var_154, x = transpose_44)[name = tensor<string, []>("input_9_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77276032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77718464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77718656)))]; | |
| tensor<fp16, [1, 77, 768]> linear_3_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_11_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; | |
| tensor<int32, [1]> input_13_axes_0 = const()[name = tensor<string, []>("input_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77720256)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77721856)))]; | |
| tensor<fp16, [1, 77, 768]> input_13_cast_fp16 = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77723456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79492992))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79493184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79495552))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_4_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")]; | |
| tensor<fp16, []> var_169_to_fp16 = const()[name = tensor<string, []>("op_169_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_170_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_169_to_fp16)[name = tensor<string, []>("op_170_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_171_cast_fp16 = sigmoid(x = var_170_cast_fp16)[name = tensor<string, []>("op_171_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_17_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_171_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79495744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81265280))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81265472)))]; | |
| tensor<fp16, [1, 77, 768]> linear_5_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_19_cast_fp16 = add(x = input_11_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81267072)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81268672)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81270272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81712704))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81712896)))]; | |
| tensor<fp16, [1, 77, 768]> linear_6_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")]; | |
| tensor<fp16, []> var_196_to_fp16 = const()[name = tensor<string, []>("op_196_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_11_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_196_to_fp16)[name = tensor<string, []>("tensor_11_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81714496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82156928))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82157120)))]; | |
| tensor<fp16, [1, 77, 768]> linear_7_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")]; | |
| tensor<int32, [4]> var_201 = const()[name = tensor<string, []>("op_201"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_202_cast_fp16 = reshape(shape = var_201, x = linear_7_cast_fp16)[name = tensor<string, []>("op_202_cast_fp16")]; | |
| tensor<int32, [4]> var_203_perm_0 = const()[name = tensor<string, []>("op_203_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82158720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82601152))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82601344)))]; | |
| tensor<fp16, [1, 77, 768]> linear_8_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")]; | |
| tensor<int32, [4]> var_208 = const()[name = tensor<string, []>("op_208"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_209_cast_fp16 = reshape(shape = var_208, x = linear_8_cast_fp16)[name = tensor<string, []>("op_209_cast_fp16")]; | |
| tensor<int32, [4]> var_210_perm_0 = const()[name = tensor<string, []>("op_210_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_217 = const()[name = tensor<string, []>("op_217"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_218_cast_fp16 = reshape(shape = var_217, x = tensor_11_cast_fp16)[name = tensor<string, []>("op_218_cast_fp16")]; | |
| tensor<int32, [4]> var_219_perm_0 = const()[name = tensor<string, []>("op_219_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_221 = const()[name = tensor<string, []>("op_221"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_43 = transpose(perm = var_219_perm_0, x = var_218_cast_fp16)[name = tensor<string, []>("transpose_43")]; | |
| tensor<fp16, [12, 77, 64]> query_states_3_cast_fp16 = reshape(shape = var_221, x = transpose_43)[name = tensor<string, []>("query_states_3_cast_fp16")]; | |
| tensor<int32, [3]> var_223 = const()[name = tensor<string, []>("op_223"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_42 = transpose(perm = var_203_perm_0, x = var_202_cast_fp16)[name = tensor<string, []>("transpose_42")]; | |
| tensor<fp16, [12, 77, 64]> key_states_7_cast_fp16 = reshape(shape = var_223, x = transpose_42)[name = tensor<string, []>("key_states_7_cast_fp16")]; | |
| tensor<int32, [3]> var_225 = const()[name = tensor<string, []>("op_225"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_41 = transpose(perm = var_210_perm_0, x = var_209_cast_fp16)[name = tensor<string, []>("transpose_41")]; | |
| tensor<fp16, [12, 77, 64]> value_states_7_cast_fp16 = reshape(shape = var_225, x = transpose_41)[name = tensor<string, []>("value_states_7_cast_fp16")]; | |
| tensor<bool, []> attn_weights_7_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_7_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_7_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_7_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_7_cast_fp16 = matmul(transpose_x = attn_weights_7_transpose_x_1, transpose_y = attn_weights_7_transpose_y_1, x = query_states_3_cast_fp16, y = key_states_7_cast_fp16)[name = tensor<string, []>("attn_weights_7_cast_fp16")]; | |
| tensor<int32, [4]> var_230 = const()[name = tensor<string, []>("op_230"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_231_cast_fp16 = reshape(shape = var_230, x = attn_weights_7_cast_fp16)[name = tensor<string, []>("op_231_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_9_cast_fp16 = add(x = var_231_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_9_cast_fp16")]; | |
| tensor<int32, [3]> var_236 = const()[name = tensor<string, []>("op_236"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_21_cast_fp16 = reshape(shape = var_236, x = attn_weights_9_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_23_cast_fp16 = softmax(axis = var_5, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; | |
| tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23_cast_fp16, y = value_states_7_cast_fp16)[name = tensor<string, []>("attn_output_7_cast_fp16")]; | |
| tensor<int32, [4]> var_241 = const()[name = tensor<string, []>("op_241"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_9_cast_fp16 = reshape(shape = var_241, x = attn_output_7_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_11_perm_0 = const()[name = tensor<string, []>("attn_output_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_244 = const()[name = tensor<string, []>("op_244"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_40 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_40")]; | |
| tensor<fp16, [1, 77, 768]> input_25_cast_fp16 = reshape(shape = var_244, x = transpose_40)[name = tensor<string, []>("input_25_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82602944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83045376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83045568)))]; | |
| tensor<fp16, [1, 77, 768]> linear_9_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; | |
| tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83047168)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83048768)))]; | |
| tensor<fp16, [1, 77, 768]> input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83050368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84819904))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84820096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84822464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_10_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")]; | |
| tensor<fp16, []> var_259_to_fp16 = const()[name = tensor<string, []>("op_259_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_260_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_259_to_fp16)[name = tensor<string, []>("op_260_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_261_cast_fp16 = sigmoid(x = var_260_cast_fp16)[name = tensor<string, []>("op_261_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_33_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_261_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84822656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86592192))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86592384)))]; | |
| tensor<fp16, [1, 77, 768]> linear_11_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_35_cast_fp16 = add(x = input_27_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = tensor<string, []>("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86593984)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86595584)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86597184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87039616))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87039808)))]; | |
| tensor<fp16, [1, 77, 768]> linear_12_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")]; | |
| tensor<fp16, []> var_286_to_fp16 = const()[name = tensor<string, []>("op_286_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_17_cast_fp16 = mul(x = linear_12_cast_fp16, y = var_286_to_fp16)[name = tensor<string, []>("tensor_17_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87041408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87483840))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87484032)))]; | |
| tensor<fp16, [1, 77, 768]> linear_13_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")]; | |
| tensor<int32, [4]> var_291 = const()[name = tensor<string, []>("op_291"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_292_cast_fp16 = reshape(shape = var_291, x = linear_13_cast_fp16)[name = tensor<string, []>("op_292_cast_fp16")]; | |
| tensor<int32, [4]> var_293_perm_0 = const()[name = tensor<string, []>("op_293_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87485632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87928064))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87928256)))]; | |
| tensor<fp16, [1, 77, 768]> linear_14_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")]; | |
| tensor<int32, [4]> var_298 = const()[name = tensor<string, []>("op_298"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_299_cast_fp16 = reshape(shape = var_298, x = linear_14_cast_fp16)[name = tensor<string, []>("op_299_cast_fp16")]; | |
| tensor<int32, [4]> var_300_perm_0 = const()[name = tensor<string, []>("op_300_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_307 = const()[name = tensor<string, []>("op_307"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_308_cast_fp16 = reshape(shape = var_307, x = tensor_17_cast_fp16)[name = tensor<string, []>("op_308_cast_fp16")]; | |
| tensor<int32, [4]> var_309_perm_0 = const()[name = tensor<string, []>("op_309_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_311 = const()[name = tensor<string, []>("op_311"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_39 = transpose(perm = var_309_perm_0, x = var_308_cast_fp16)[name = tensor<string, []>("transpose_39")]; | |
| tensor<fp16, [12, 77, 64]> query_states_5_cast_fp16 = reshape(shape = var_311, x = transpose_39)[name = tensor<string, []>("query_states_5_cast_fp16")]; | |
| tensor<int32, [3]> var_313 = const()[name = tensor<string, []>("op_313"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_38 = transpose(perm = var_293_perm_0, x = var_292_cast_fp16)[name = tensor<string, []>("transpose_38")]; | |
| tensor<fp16, [12, 77, 64]> key_states_11_cast_fp16 = reshape(shape = var_313, x = transpose_38)[name = tensor<string, []>("key_states_11_cast_fp16")]; | |
| tensor<int32, [3]> var_315 = const()[name = tensor<string, []>("op_315"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_37 = transpose(perm = var_300_perm_0, x = var_299_cast_fp16)[name = tensor<string, []>("transpose_37")]; | |
| tensor<fp16, [12, 77, 64]> value_states_11_cast_fp16 = reshape(shape = var_315, x = transpose_37)[name = tensor<string, []>("value_states_11_cast_fp16")]; | |
| tensor<bool, []> attn_weights_13_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_13_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_13_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_13_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_13_cast_fp16 = matmul(transpose_x = attn_weights_13_transpose_x_1, transpose_y = attn_weights_13_transpose_y_1, x = query_states_5_cast_fp16, y = key_states_11_cast_fp16)[name = tensor<string, []>("attn_weights_13_cast_fp16")]; | |
| tensor<int32, [4]> var_320 = const()[name = tensor<string, []>("op_320"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_321_cast_fp16 = reshape(shape = var_320, x = attn_weights_13_cast_fp16)[name = tensor<string, []>("op_321_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_15_cast_fp16 = add(x = var_321_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_15_cast_fp16")]; | |
| tensor<int32, [3]> var_326 = const()[name = tensor<string, []>("op_326"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_37_cast_fp16 = reshape(shape = var_326, x = attn_weights_15_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_39_cast_fp16 = softmax(axis = var_5, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; | |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39_cast_fp16, y = value_states_11_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")]; | |
| tensor<int32, [4]> var_331 = const()[name = tensor<string, []>("op_331"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_15_cast_fp16 = reshape(shape = var_331, x = attn_output_13_cast_fp16)[name = tensor<string, []>("attn_output_15_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_17_perm_0 = const()[name = tensor<string, []>("attn_output_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_334 = const()[name = tensor<string, []>("op_334"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_36 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast_fp16)[name = tensor<string, []>("transpose_36")]; | |
| tensor<fp16, [1, 77, 768]> input_41_cast_fp16 = reshape(shape = var_334, x = transpose_36)[name = tensor<string, []>("input_41_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87929856))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88372288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88372480)))]; | |
| tensor<fp16, [1, 77, 768]> linear_15_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_43_cast_fp16 = add(x = input_35_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; | |
| tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88374080)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88375680)))]; | |
| tensor<fp16, [1, 77, 768]> input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88377280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90146816))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90147008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90149376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_16_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")]; | |
| tensor<fp16, []> var_349_to_fp16 = const()[name = tensor<string, []>("op_349_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_350_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_349_to_fp16)[name = tensor<string, []>("op_350_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_351_cast_fp16 = sigmoid(x = var_350_cast_fp16)[name = tensor<string, []>("op_351_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_49_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_351_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90149568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91919104))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91919296)))]; | |
| tensor<fp16, [1, 77, 768]> linear_17_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = tensor<string, []>("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91920896)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91922496)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91924096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92366528))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92366720)))]; | |
| tensor<fp16, [1, 77, 768]> linear_18_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")]; | |
| tensor<fp16, []> var_376_to_fp16 = const()[name = tensor<string, []>("op_376_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_23_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_376_to_fp16)[name = tensor<string, []>("tensor_23_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92368320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92810752))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92810944)))]; | |
| tensor<fp16, [1, 77, 768]> linear_19_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")]; | |
| tensor<int32, [4]> var_381 = const()[name = tensor<string, []>("op_381"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_382_cast_fp16 = reshape(shape = var_381, x = linear_19_cast_fp16)[name = tensor<string, []>("op_382_cast_fp16")]; | |
| tensor<int32, [4]> var_383_perm_0 = const()[name = tensor<string, []>("op_383_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92812544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93254976))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93255168)))]; | |
| tensor<fp16, [1, 77, 768]> linear_20_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")]; | |
| tensor<int32, [4]> var_388 = const()[name = tensor<string, []>("op_388"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_389_cast_fp16 = reshape(shape = var_388, x = linear_20_cast_fp16)[name = tensor<string, []>("op_389_cast_fp16")]; | |
| tensor<int32, [4]> var_390_perm_0 = const()[name = tensor<string, []>("op_390_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_397 = const()[name = tensor<string, []>("op_397"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_398_cast_fp16 = reshape(shape = var_397, x = tensor_23_cast_fp16)[name = tensor<string, []>("op_398_cast_fp16")]; | |
| tensor<int32, [4]> var_399_perm_0 = const()[name = tensor<string, []>("op_399_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_401 = const()[name = tensor<string, []>("op_401"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_35 = transpose(perm = var_399_perm_0, x = var_398_cast_fp16)[name = tensor<string, []>("transpose_35")]; | |
| tensor<fp16, [12, 77, 64]> query_states_7_cast_fp16 = reshape(shape = var_401, x = transpose_35)[name = tensor<string, []>("query_states_7_cast_fp16")]; | |
| tensor<int32, [3]> var_403 = const()[name = tensor<string, []>("op_403"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_34 = transpose(perm = var_383_perm_0, x = var_382_cast_fp16)[name = tensor<string, []>("transpose_34")]; | |
| tensor<fp16, [12, 77, 64]> key_states_15_cast_fp16 = reshape(shape = var_403, x = transpose_34)[name = tensor<string, []>("key_states_15_cast_fp16")]; | |
| tensor<int32, [3]> var_405 = const()[name = tensor<string, []>("op_405"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_33 = transpose(perm = var_390_perm_0, x = var_389_cast_fp16)[name = tensor<string, []>("transpose_33")]; | |
| tensor<fp16, [12, 77, 64]> value_states_15_cast_fp16 = reshape(shape = var_405, x = transpose_33)[name = tensor<string, []>("value_states_15_cast_fp16")]; | |
| tensor<bool, []> attn_weights_19_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_19_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_19_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_19_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_19_cast_fp16 = matmul(transpose_x = attn_weights_19_transpose_x_1, transpose_y = attn_weights_19_transpose_y_1, x = query_states_7_cast_fp16, y = key_states_15_cast_fp16)[name = tensor<string, []>("attn_weights_19_cast_fp16")]; | |
| tensor<int32, [4]> var_410 = const()[name = tensor<string, []>("op_410"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_411_cast_fp16 = reshape(shape = var_410, x = attn_weights_19_cast_fp16)[name = tensor<string, []>("op_411_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_21_cast_fp16 = add(x = var_411_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_21_cast_fp16")]; | |
| tensor<int32, [3]> var_416 = const()[name = tensor<string, []>("op_416"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_53_cast_fp16 = reshape(shape = var_416, x = attn_weights_21_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_55_cast_fp16 = softmax(axis = var_5, x = input_53_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; | |
| tensor<bool, []> attn_output_19_transpose_x_0 = const()[name = tensor<string, []>("attn_output_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_19_transpose_y_0 = const()[name = tensor<string, []>("attn_output_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_19_cast_fp16 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55_cast_fp16, y = value_states_15_cast_fp16)[name = tensor<string, []>("attn_output_19_cast_fp16")]; | |
| tensor<int32, [4]> var_421 = const()[name = tensor<string, []>("op_421"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_21_cast_fp16 = reshape(shape = var_421, x = attn_output_19_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_23_perm_0 = const()[name = tensor<string, []>("attn_output_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_424 = const()[name = tensor<string, []>("op_424"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_32 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_32")]; | |
| tensor<fp16, [1, 77, 768]> input_57_cast_fp16 = reshape(shape = var_424, x = transpose_32)[name = tensor<string, []>("input_57_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93256768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93699200))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93699392)))]; | |
| tensor<fp16, [1, 77, 768]> linear_21_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_59_cast_fp16 = add(x = input_51_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; | |
| tensor<int32, [1]> input_61_axes_0 = const()[name = tensor<string, []>("input_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93700992)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93702592)))]; | |
| tensor<fp16, [1, 77, 768]> input_61_cast_fp16 = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93704192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95473728))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95473920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95476288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_22_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")]; | |
| tensor<fp16, []> var_439_to_fp16 = const()[name = tensor<string, []>("op_439_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_440_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_439_to_fp16)[name = tensor<string, []>("op_440_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_441_cast_fp16 = sigmoid(x = var_440_cast_fp16)[name = tensor<string, []>("op_441_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_65_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_441_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95476480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97246016))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97246208)))]; | |
| tensor<fp16, [1, 77, 768]> linear_23_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_67_cast_fp16 = add(x = input_59_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97247808)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97249408)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("hidden_states_25_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97251008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97693440))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97693632)))]; | |
| tensor<fp16, [1, 77, 768]> linear_24_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")]; | |
| tensor<fp16, []> var_466_to_fp16 = const()[name = tensor<string, []>("op_466_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_29_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_466_to_fp16)[name = tensor<string, []>("tensor_29_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97695232))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98137664))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98137856)))]; | |
| tensor<fp16, [1, 77, 768]> linear_25_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")]; | |
| tensor<int32, [4]> var_471 = const()[name = tensor<string, []>("op_471"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_472_cast_fp16 = reshape(shape = var_471, x = linear_25_cast_fp16)[name = tensor<string, []>("op_472_cast_fp16")]; | |
| tensor<int32, [4]> var_473_perm_0 = const()[name = tensor<string, []>("op_473_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98139456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98581888))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98582080)))]; | |
| tensor<fp16, [1, 77, 768]> linear_26_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")]; | |
| tensor<int32, [4]> var_478 = const()[name = tensor<string, []>("op_478"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_479_cast_fp16 = reshape(shape = var_478, x = linear_26_cast_fp16)[name = tensor<string, []>("op_479_cast_fp16")]; | |
| tensor<int32, [4]> var_480_perm_0 = const()[name = tensor<string, []>("op_480_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_487 = const()[name = tensor<string, []>("op_487"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_488_cast_fp16 = reshape(shape = var_487, x = tensor_29_cast_fp16)[name = tensor<string, []>("op_488_cast_fp16")]; | |
| tensor<int32, [4]> var_489_perm_0 = const()[name = tensor<string, []>("op_489_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_491 = const()[name = tensor<string, []>("op_491"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_31 = transpose(perm = var_489_perm_0, x = var_488_cast_fp16)[name = tensor<string, []>("transpose_31")]; | |
| tensor<fp16, [12, 77, 64]> query_states_9_cast_fp16 = reshape(shape = var_491, x = transpose_31)[name = tensor<string, []>("query_states_9_cast_fp16")]; | |
| tensor<int32, [3]> var_493 = const()[name = tensor<string, []>("op_493"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_30 = transpose(perm = var_473_perm_0, x = var_472_cast_fp16)[name = tensor<string, []>("transpose_30")]; | |
| tensor<fp16, [12, 77, 64]> key_states_19_cast_fp16 = reshape(shape = var_493, x = transpose_30)[name = tensor<string, []>("key_states_19_cast_fp16")]; | |
| tensor<int32, [3]> var_495 = const()[name = tensor<string, []>("op_495"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_29 = transpose(perm = var_480_perm_0, x = var_479_cast_fp16)[name = tensor<string, []>("transpose_29")]; | |
| tensor<fp16, [12, 77, 64]> value_states_19_cast_fp16 = reshape(shape = var_495, x = transpose_29)[name = tensor<string, []>("value_states_19_cast_fp16")]; | |
| tensor<bool, []> attn_weights_25_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_25_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_25_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_25_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_25_cast_fp16 = matmul(transpose_x = attn_weights_25_transpose_x_1, transpose_y = attn_weights_25_transpose_y_1, x = query_states_9_cast_fp16, y = key_states_19_cast_fp16)[name = tensor<string, []>("attn_weights_25_cast_fp16")]; | |
| tensor<int32, [4]> var_500 = const()[name = tensor<string, []>("op_500"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_501_cast_fp16 = reshape(shape = var_500, x = attn_weights_25_cast_fp16)[name = tensor<string, []>("op_501_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_27_cast_fp16 = add(x = var_501_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_27_cast_fp16")]; | |
| tensor<int32, [3]> var_506 = const()[name = tensor<string, []>("op_506"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_69_cast_fp16 = reshape(shape = var_506, x = attn_weights_27_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_71_cast_fp16 = softmax(axis = var_5, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; | |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71_cast_fp16, y = value_states_19_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")]; | |
| tensor<int32, [4]> var_511 = const()[name = tensor<string, []>("op_511"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_27_cast_fp16 = reshape(shape = var_511, x = attn_output_25_cast_fp16)[name = tensor<string, []>("attn_output_27_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_29_perm_0 = const()[name = tensor<string, []>("attn_output_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_514 = const()[name = tensor<string, []>("op_514"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_28 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast_fp16)[name = tensor<string, []>("transpose_28")]; | |
| tensor<fp16, [1, 77, 768]> input_73_cast_fp16 = reshape(shape = var_514, x = transpose_28)[name = tensor<string, []>("input_73_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98583680))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99026112))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99026304)))]; | |
| tensor<fp16, [1, 77, 768]> linear_27_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")]; | |
| tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99027904)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99029504)))]; | |
| tensor<fp16, [1, 77, 768]> input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99031104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100800640))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100800832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100803200))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_28_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")]; | |
| tensor<fp16, []> var_529_to_fp16 = const()[name = tensor<string, []>("op_529_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_530_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_529_to_fp16)[name = tensor<string, []>("op_530_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_531_cast_fp16 = sigmoid(x = var_530_cast_fp16)[name = tensor<string, []>("op_531_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_81_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_531_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100803392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102572928))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102573120)))]; | |
| tensor<fp16, [1, 77, 768]> linear_29_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_83_cast_fp16 = add(x = input_75_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102574720)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102576320)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("hidden_states_31_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102577920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103020352))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103020544)))]; | |
| tensor<fp16, [1, 77, 768]> linear_30_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")]; | |
| tensor<fp16, []> var_556_to_fp16 = const()[name = tensor<string, []>("op_556_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_35_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_556_to_fp16)[name = tensor<string, []>("tensor_35_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103022144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103464576))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103464768)))]; | |
| tensor<fp16, [1, 77, 768]> linear_31_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")]; | |
| tensor<int32, [4]> var_561 = const()[name = tensor<string, []>("op_561"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_562_cast_fp16 = reshape(shape = var_561, x = linear_31_cast_fp16)[name = tensor<string, []>("op_562_cast_fp16")]; | |
| tensor<int32, [4]> var_563_perm_0 = const()[name = tensor<string, []>("op_563_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103466368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103908800))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103908992)))]; | |
| tensor<fp16, [1, 77, 768]> linear_32_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")]; | |
| tensor<int32, [4]> var_568 = const()[name = tensor<string, []>("op_568"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_569_cast_fp16 = reshape(shape = var_568, x = linear_32_cast_fp16)[name = tensor<string, []>("op_569_cast_fp16")]; | |
| tensor<int32, [4]> var_570_perm_0 = const()[name = tensor<string, []>("op_570_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_577 = const()[name = tensor<string, []>("op_577"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_578_cast_fp16 = reshape(shape = var_577, x = tensor_35_cast_fp16)[name = tensor<string, []>("op_578_cast_fp16")]; | |
| tensor<int32, [4]> var_579_perm_0 = const()[name = tensor<string, []>("op_579_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_581 = const()[name = tensor<string, []>("op_581"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_27 = transpose(perm = var_579_perm_0, x = var_578_cast_fp16)[name = tensor<string, []>("transpose_27")]; | |
| tensor<fp16, [12, 77, 64]> query_states_11_cast_fp16 = reshape(shape = var_581, x = transpose_27)[name = tensor<string, []>("query_states_11_cast_fp16")]; | |
| tensor<int32, [3]> var_583 = const()[name = tensor<string, []>("op_583"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_26 = transpose(perm = var_563_perm_0, x = var_562_cast_fp16)[name = tensor<string, []>("transpose_26")]; | |
| tensor<fp16, [12, 77, 64]> key_states_23_cast_fp16 = reshape(shape = var_583, x = transpose_26)[name = tensor<string, []>("key_states_23_cast_fp16")]; | |
| tensor<int32, [3]> var_585 = const()[name = tensor<string, []>("op_585"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_25 = transpose(perm = var_570_perm_0, x = var_569_cast_fp16)[name = tensor<string, []>("transpose_25")]; | |
| tensor<fp16, [12, 77, 64]> value_states_23_cast_fp16 = reshape(shape = var_585, x = transpose_25)[name = tensor<string, []>("value_states_23_cast_fp16")]; | |
| tensor<bool, []> attn_weights_31_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_31_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_31_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_31_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_31_cast_fp16 = matmul(transpose_x = attn_weights_31_transpose_x_1, transpose_y = attn_weights_31_transpose_y_1, x = query_states_11_cast_fp16, y = key_states_23_cast_fp16)[name = tensor<string, []>("attn_weights_31_cast_fp16")]; | |
| tensor<int32, [4]> var_590 = const()[name = tensor<string, []>("op_590"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_591_cast_fp16 = reshape(shape = var_590, x = attn_weights_31_cast_fp16)[name = tensor<string, []>("op_591_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_33_cast_fp16 = add(x = var_591_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_33_cast_fp16")]; | |
| tensor<int32, [3]> var_596 = const()[name = tensor<string, []>("op_596"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_85_cast_fp16 = reshape(shape = var_596, x = attn_weights_33_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_87_cast_fp16 = softmax(axis = var_5, x = input_85_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; | |
| tensor<bool, []> attn_output_31_transpose_x_0 = const()[name = tensor<string, []>("attn_output_31_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_31_transpose_y_0 = const()[name = tensor<string, []>("attn_output_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_31_cast_fp16 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87_cast_fp16, y = value_states_23_cast_fp16)[name = tensor<string, []>("attn_output_31_cast_fp16")]; | |
| tensor<int32, [4]> var_601 = const()[name = tensor<string, []>("op_601"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_33_cast_fp16 = reshape(shape = var_601, x = attn_output_31_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_35_perm_0 = const()[name = tensor<string, []>("attn_output_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_604 = const()[name = tensor<string, []>("op_604"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_24 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_24")]; | |
| tensor<fp16, [1, 77, 768]> input_89_cast_fp16 = reshape(shape = var_604, x = transpose_24)[name = tensor<string, []>("input_89_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103910592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104353024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104353216)))]; | |
| tensor<fp16, [1, 77, 768]> linear_33_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; | |
| tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104354816)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104356416)))]; | |
| tensor<fp16, [1, 77, 768]> input_93_cast_fp16 = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104358016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106127552))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106127744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106130112))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_34_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")]; | |
| tensor<fp16, []> var_619_to_fp16 = const()[name = tensor<string, []>("op_619_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_620_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_619_to_fp16)[name = tensor<string, []>("op_620_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_621_cast_fp16 = sigmoid(x = var_620_cast_fp16)[name = tensor<string, []>("op_621_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_97_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_621_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106130304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107899840))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107900032)))]; | |
| tensor<fp16, [1, 77, 768]> linear_35_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = tensor<string, []>("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107901632)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107903232)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("hidden_states_37_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107904832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108347264))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108347456)))]; | |
| tensor<fp16, [1, 77, 768]> linear_36_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")]; | |
| tensor<fp16, []> var_646_to_fp16 = const()[name = tensor<string, []>("op_646_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_41_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_646_to_fp16)[name = tensor<string, []>("tensor_41_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108349056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108791488))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108791680)))]; | |
| tensor<fp16, [1, 77, 768]> linear_37_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")]; | |
| tensor<int32, [4]> var_651 = const()[name = tensor<string, []>("op_651"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_652_cast_fp16 = reshape(shape = var_651, x = linear_37_cast_fp16)[name = tensor<string, []>("op_652_cast_fp16")]; | |
| tensor<int32, [4]> var_653_perm_0 = const()[name = tensor<string, []>("op_653_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108793280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109235712))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109235904)))]; | |
| tensor<fp16, [1, 77, 768]> linear_38_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")]; | |
| tensor<int32, [4]> var_658 = const()[name = tensor<string, []>("op_658"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_659_cast_fp16 = reshape(shape = var_658, x = linear_38_cast_fp16)[name = tensor<string, []>("op_659_cast_fp16")]; | |
| tensor<int32, [4]> var_660_perm_0 = const()[name = tensor<string, []>("op_660_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_667 = const()[name = tensor<string, []>("op_667"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_668_cast_fp16 = reshape(shape = var_667, x = tensor_41_cast_fp16)[name = tensor<string, []>("op_668_cast_fp16")]; | |
| tensor<int32, [4]> var_669_perm_0 = const()[name = tensor<string, []>("op_669_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_671 = const()[name = tensor<string, []>("op_671"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_23 = transpose(perm = var_669_perm_0, x = var_668_cast_fp16)[name = tensor<string, []>("transpose_23")]; | |
| tensor<fp16, [12, 77, 64]> query_states_13_cast_fp16 = reshape(shape = var_671, x = transpose_23)[name = tensor<string, []>("query_states_13_cast_fp16")]; | |
| tensor<int32, [3]> var_673 = const()[name = tensor<string, []>("op_673"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_22 = transpose(perm = var_653_perm_0, x = var_652_cast_fp16)[name = tensor<string, []>("transpose_22")]; | |
| tensor<fp16, [12, 77, 64]> key_states_27_cast_fp16 = reshape(shape = var_673, x = transpose_22)[name = tensor<string, []>("key_states_27_cast_fp16")]; | |
| tensor<int32, [3]> var_675 = const()[name = tensor<string, []>("op_675"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_21 = transpose(perm = var_660_perm_0, x = var_659_cast_fp16)[name = tensor<string, []>("transpose_21")]; | |
| tensor<fp16, [12, 77, 64]> value_states_27_cast_fp16 = reshape(shape = var_675, x = transpose_21)[name = tensor<string, []>("value_states_27_cast_fp16")]; | |
| tensor<bool, []> attn_weights_37_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_37_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_37_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_37_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_37_cast_fp16 = matmul(transpose_x = attn_weights_37_transpose_x_1, transpose_y = attn_weights_37_transpose_y_1, x = query_states_13_cast_fp16, y = key_states_27_cast_fp16)[name = tensor<string, []>("attn_weights_37_cast_fp16")]; | |
| tensor<int32, [4]> var_680 = const()[name = tensor<string, []>("op_680"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_681_cast_fp16 = reshape(shape = var_680, x = attn_weights_37_cast_fp16)[name = tensor<string, []>("op_681_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_39_cast_fp16 = add(x = var_681_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_39_cast_fp16")]; | |
| tensor<int32, [3]> var_686 = const()[name = tensor<string, []>("op_686"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_101_cast_fp16 = reshape(shape = var_686, x = attn_weights_39_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_103_cast_fp16 = softmax(axis = var_5, x = input_101_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; | |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103_cast_fp16, y = value_states_27_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")]; | |
| tensor<int32, [4]> var_691 = const()[name = tensor<string, []>("op_691"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_39_cast_fp16 = reshape(shape = var_691, x = attn_output_37_cast_fp16)[name = tensor<string, []>("attn_output_39_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_41_perm_0 = const()[name = tensor<string, []>("attn_output_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_694 = const()[name = tensor<string, []>("op_694"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_20 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast_fp16)[name = tensor<string, []>("transpose_20")]; | |
| tensor<fp16, [1, 77, 768]> input_105_cast_fp16 = reshape(shape = var_694, x = transpose_20)[name = tensor<string, []>("input_105_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109237504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109679936))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109680128)))]; | |
| tensor<fp16, [1, 77, 768]> linear_39_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_107_cast_fp16 = add(x = input_99_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")]; | |
| tensor<int32, [1]> input_109_axes_0 = const()[name = tensor<string, []>("input_109_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109681728)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109683328)))]; | |
| tensor<fp16, [1, 77, 768]> input_109_cast_fp16 = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_107_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109684928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111454464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111454656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111457024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_40_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")]; | |
| tensor<fp16, []> var_709_to_fp16 = const()[name = tensor<string, []>("op_709_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_710_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_709_to_fp16)[name = tensor<string, []>("op_710_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_711_cast_fp16 = sigmoid(x = var_710_cast_fp16)[name = tensor<string, []>("op_711_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_113_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_711_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111457216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113226752))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113226944)))]; | |
| tensor<fp16, [1, 77, 768]> linear_41_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_115_cast_fp16 = add(x = input_107_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113228544)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113230144)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("hidden_states_43_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113231744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113674176))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113674368)))]; | |
| tensor<fp16, [1, 77, 768]> linear_42_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")]; | |
| tensor<fp16, []> var_736_to_fp16 = const()[name = tensor<string, []>("op_736_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_47_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_736_to_fp16)[name = tensor<string, []>("tensor_47_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113675968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114118400))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114118592)))]; | |
| tensor<fp16, [1, 77, 768]> linear_43_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")]; | |
| tensor<int32, [4]> var_741 = const()[name = tensor<string, []>("op_741"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_742_cast_fp16 = reshape(shape = var_741, x = linear_43_cast_fp16)[name = tensor<string, []>("op_742_cast_fp16")]; | |
| tensor<int32, [4]> var_743_perm_0 = const()[name = tensor<string, []>("op_743_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114120192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114562624))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114562816)))]; | |
| tensor<fp16, [1, 77, 768]> linear_44_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")]; | |
| tensor<int32, [4]> var_748 = const()[name = tensor<string, []>("op_748"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_749_cast_fp16 = reshape(shape = var_748, x = linear_44_cast_fp16)[name = tensor<string, []>("op_749_cast_fp16")]; | |
| tensor<int32, [4]> var_750_perm_0 = const()[name = tensor<string, []>("op_750_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_757 = const()[name = tensor<string, []>("op_757"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_758_cast_fp16 = reshape(shape = var_757, x = tensor_47_cast_fp16)[name = tensor<string, []>("op_758_cast_fp16")]; | |
| tensor<int32, [4]> var_759_perm_0 = const()[name = tensor<string, []>("op_759_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_761 = const()[name = tensor<string, []>("op_761"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_19 = transpose(perm = var_759_perm_0, x = var_758_cast_fp16)[name = tensor<string, []>("transpose_19")]; | |
| tensor<fp16, [12, 77, 64]> query_states_15_cast_fp16 = reshape(shape = var_761, x = transpose_19)[name = tensor<string, []>("query_states_15_cast_fp16")]; | |
| tensor<int32, [3]> var_763 = const()[name = tensor<string, []>("op_763"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_18 = transpose(perm = var_743_perm_0, x = var_742_cast_fp16)[name = tensor<string, []>("transpose_18")]; | |
| tensor<fp16, [12, 77, 64]> key_states_31_cast_fp16 = reshape(shape = var_763, x = transpose_18)[name = tensor<string, []>("key_states_31_cast_fp16")]; | |
| tensor<int32, [3]> var_765 = const()[name = tensor<string, []>("op_765"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_17 = transpose(perm = var_750_perm_0, x = var_749_cast_fp16)[name = tensor<string, []>("transpose_17")]; | |
| tensor<fp16, [12, 77, 64]> value_states_31_cast_fp16 = reshape(shape = var_765, x = transpose_17)[name = tensor<string, []>("value_states_31_cast_fp16")]; | |
| tensor<bool, []> attn_weights_43_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_43_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_43_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_43_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_43_cast_fp16 = matmul(transpose_x = attn_weights_43_transpose_x_1, transpose_y = attn_weights_43_transpose_y_1, x = query_states_15_cast_fp16, y = key_states_31_cast_fp16)[name = tensor<string, []>("attn_weights_43_cast_fp16")]; | |
| tensor<int32, [4]> var_770 = const()[name = tensor<string, []>("op_770"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_771_cast_fp16 = reshape(shape = var_770, x = attn_weights_43_cast_fp16)[name = tensor<string, []>("op_771_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_45_cast_fp16 = add(x = var_771_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_45_cast_fp16")]; | |
| tensor<int32, [3]> var_776 = const()[name = tensor<string, []>("op_776"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_117_cast_fp16 = reshape(shape = var_776, x = attn_weights_45_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_119_cast_fp16 = softmax(axis = var_5, x = input_117_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")]; | |
| tensor<bool, []> attn_output_43_transpose_x_0 = const()[name = tensor<string, []>("attn_output_43_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_43_transpose_y_0 = const()[name = tensor<string, []>("attn_output_43_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_43_cast_fp16 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119_cast_fp16, y = value_states_31_cast_fp16)[name = tensor<string, []>("attn_output_43_cast_fp16")]; | |
| tensor<int32, [4]> var_781 = const()[name = tensor<string, []>("op_781"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_45_cast_fp16 = reshape(shape = var_781, x = attn_output_43_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_47_perm_0 = const()[name = tensor<string, []>("attn_output_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_784 = const()[name = tensor<string, []>("op_784"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_16 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_16")]; | |
| tensor<fp16, [1, 77, 768]> input_121_cast_fp16 = reshape(shape = var_784, x = transpose_16)[name = tensor<string, []>("input_121_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114564416))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115006848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115007040)))]; | |
| tensor<fp16, [1, 77, 768]> linear_45_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")]; | |
| tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115008640)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115010240)))]; | |
| tensor<fp16, [1, 77, 768]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115011840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116781376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116781568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116783936))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_46_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")]; | |
| tensor<fp16, []> var_799_to_fp16 = const()[name = tensor<string, []>("op_799_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_800_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_799_to_fp16)[name = tensor<string, []>("op_800_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_801_cast_fp16 = sigmoid(x = var_800_cast_fp16)[name = tensor<string, []>("op_801_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_129_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_801_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116784128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118553664))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118553856)))]; | |
| tensor<fp16, [1, 77, 768]> linear_47_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_131_cast_fp16 = add(x = input_123_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("input_131_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = tensor<string, []>("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118555456)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118557056)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_131_cast_fp16)[name = tensor<string, []>("hidden_states_49_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118558656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119001088))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119001280)))]; | |
| tensor<fp16, [1, 77, 768]> linear_48_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")]; | |
| tensor<fp16, []> var_826_to_fp16 = const()[name = tensor<string, []>("op_826_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_53_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_826_to_fp16)[name = tensor<string, []>("tensor_53_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119002880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119445312))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119445504)))]; | |
| tensor<fp16, [1, 77, 768]> linear_49_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")]; | |
| tensor<int32, [4]> var_831 = const()[name = tensor<string, []>("op_831"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_832_cast_fp16 = reshape(shape = var_831, x = linear_49_cast_fp16)[name = tensor<string, []>("op_832_cast_fp16")]; | |
| tensor<int32, [4]> var_833_perm_0 = const()[name = tensor<string, []>("op_833_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119447104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119889536))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119889728)))]; | |
| tensor<fp16, [1, 77, 768]> linear_50_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")]; | |
| tensor<int32, [4]> var_838 = const()[name = tensor<string, []>("op_838"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_839_cast_fp16 = reshape(shape = var_838, x = linear_50_cast_fp16)[name = tensor<string, []>("op_839_cast_fp16")]; | |
| tensor<int32, [4]> var_840_perm_0 = const()[name = tensor<string, []>("op_840_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_847 = const()[name = tensor<string, []>("op_847"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_848_cast_fp16 = reshape(shape = var_847, x = tensor_53_cast_fp16)[name = tensor<string, []>("op_848_cast_fp16")]; | |
| tensor<int32, [4]> var_849_perm_0 = const()[name = tensor<string, []>("op_849_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_851 = const()[name = tensor<string, []>("op_851"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_15 = transpose(perm = var_849_perm_0, x = var_848_cast_fp16)[name = tensor<string, []>("transpose_15")]; | |
| tensor<fp16, [12, 77, 64]> query_states_17_cast_fp16 = reshape(shape = var_851, x = transpose_15)[name = tensor<string, []>("query_states_17_cast_fp16")]; | |
| tensor<int32, [3]> var_853 = const()[name = tensor<string, []>("op_853"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_14 = transpose(perm = var_833_perm_0, x = var_832_cast_fp16)[name = tensor<string, []>("transpose_14")]; | |
| tensor<fp16, [12, 77, 64]> key_states_35_cast_fp16 = reshape(shape = var_853, x = transpose_14)[name = tensor<string, []>("key_states_35_cast_fp16")]; | |
| tensor<int32, [3]> var_855 = const()[name = tensor<string, []>("op_855"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_13 = transpose(perm = var_840_perm_0, x = var_839_cast_fp16)[name = tensor<string, []>("transpose_13")]; | |
| tensor<fp16, [12, 77, 64]> value_states_35_cast_fp16 = reshape(shape = var_855, x = transpose_13)[name = tensor<string, []>("value_states_35_cast_fp16")]; | |
| tensor<bool, []> attn_weights_49_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_49_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_49_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_49_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_49_cast_fp16 = matmul(transpose_x = attn_weights_49_transpose_x_1, transpose_y = attn_weights_49_transpose_y_1, x = query_states_17_cast_fp16, y = key_states_35_cast_fp16)[name = tensor<string, []>("attn_weights_49_cast_fp16")]; | |
| tensor<int32, [4]> var_860 = const()[name = tensor<string, []>("op_860"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_861_cast_fp16 = reshape(shape = var_860, x = attn_weights_49_cast_fp16)[name = tensor<string, []>("op_861_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_51_cast_fp16 = add(x = var_861_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_51_cast_fp16")]; | |
| tensor<int32, [3]> var_866 = const()[name = tensor<string, []>("op_866"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_133_cast_fp16 = reshape(shape = var_866, x = attn_weights_51_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_135_cast_fp16 = softmax(axis = var_5, x = input_133_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")]; | |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135_cast_fp16, y = value_states_35_cast_fp16)[name = tensor<string, []>("attn_output_49_cast_fp16")]; | |
| tensor<int32, [4]> var_871 = const()[name = tensor<string, []>("op_871"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_51_cast_fp16 = reshape(shape = var_871, x = attn_output_49_cast_fp16)[name = tensor<string, []>("attn_output_51_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_53_perm_0 = const()[name = tensor<string, []>("attn_output_53_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_874 = const()[name = tensor<string, []>("op_874"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_12 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast_fp16)[name = tensor<string, []>("transpose_12")]; | |
| tensor<fp16, [1, 77, 768]> input_137_cast_fp16 = reshape(shape = var_874, x = transpose_12)[name = tensor<string, []>("input_137_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119891328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120333760))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120333952)))]; | |
| tensor<fp16, [1, 77, 768]> linear_51_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_139_cast_fp16 = add(x = input_131_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")]; | |
| tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120335552)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120337152)))]; | |
| tensor<fp16, [1, 77, 768]> input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120338752))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122108288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122108480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122110848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_52_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")]; | |
| tensor<fp16, []> var_889_to_fp16 = const()[name = tensor<string, []>("op_889_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_890_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_889_to_fp16)[name = tensor<string, []>("op_890_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_891_cast_fp16 = sigmoid(x = var_890_cast_fp16)[name = tensor<string, []>("op_891_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_145_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_891_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122111040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123880576))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123880768)))]; | |
| tensor<fp16, [1, 77, 768]> linear_53_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123882368)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123883968)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor<string, []>("hidden_states_55_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123885568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124328000))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124328192)))]; | |
| tensor<fp16, [1, 77, 768]> linear_54_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")]; | |
| tensor<fp16, []> var_916_to_fp16 = const()[name = tensor<string, []>("op_916_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_59_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_916_to_fp16)[name = tensor<string, []>("tensor_59_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124329792))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124772224))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124772416)))]; | |
| tensor<fp16, [1, 77, 768]> linear_55_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")]; | |
| tensor<int32, [4]> var_921 = const()[name = tensor<string, []>("op_921"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_922_cast_fp16 = reshape(shape = var_921, x = linear_55_cast_fp16)[name = tensor<string, []>("op_922_cast_fp16")]; | |
| tensor<int32, [4]> var_923_perm_0 = const()[name = tensor<string, []>("op_923_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124774016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125216448))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125216640)))]; | |
| tensor<fp16, [1, 77, 768]> linear_56_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")]; | |
| tensor<int32, [4]> var_928 = const()[name = tensor<string, []>("op_928"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_929_cast_fp16 = reshape(shape = var_928, x = linear_56_cast_fp16)[name = tensor<string, []>("op_929_cast_fp16")]; | |
| tensor<int32, [4]> var_930_perm_0 = const()[name = tensor<string, []>("op_930_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_937 = const()[name = tensor<string, []>("op_937"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_938_cast_fp16 = reshape(shape = var_937, x = tensor_59_cast_fp16)[name = tensor<string, []>("op_938_cast_fp16")]; | |
| tensor<int32, [4]> var_939_perm_0 = const()[name = tensor<string, []>("op_939_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_941 = const()[name = tensor<string, []>("op_941"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_11 = transpose(perm = var_939_perm_0, x = var_938_cast_fp16)[name = tensor<string, []>("transpose_11")]; | |
| tensor<fp16, [12, 77, 64]> query_states_19_cast_fp16 = reshape(shape = var_941, x = transpose_11)[name = tensor<string, []>("query_states_19_cast_fp16")]; | |
| tensor<int32, [3]> var_943 = const()[name = tensor<string, []>("op_943"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_10 = transpose(perm = var_923_perm_0, x = var_922_cast_fp16)[name = tensor<string, []>("transpose_10")]; | |
| tensor<fp16, [12, 77, 64]> key_states_39_cast_fp16 = reshape(shape = var_943, x = transpose_10)[name = tensor<string, []>("key_states_39_cast_fp16")]; | |
| tensor<int32, [3]> var_945 = const()[name = tensor<string, []>("op_945"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_9 = transpose(perm = var_930_perm_0, x = var_929_cast_fp16)[name = tensor<string, []>("transpose_9")]; | |
| tensor<fp16, [12, 77, 64]> value_states_39_cast_fp16 = reshape(shape = var_945, x = transpose_9)[name = tensor<string, []>("value_states_39_cast_fp16")]; | |
| tensor<bool, []> attn_weights_55_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_55_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_55_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_55_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_55_cast_fp16 = matmul(transpose_x = attn_weights_55_transpose_x_1, transpose_y = attn_weights_55_transpose_y_1, x = query_states_19_cast_fp16, y = key_states_39_cast_fp16)[name = tensor<string, []>("attn_weights_55_cast_fp16")]; | |
| tensor<int32, [4]> var_950 = const()[name = tensor<string, []>("op_950"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_951_cast_fp16 = reshape(shape = var_950, x = attn_weights_55_cast_fp16)[name = tensor<string, []>("op_951_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_57_cast_fp16 = add(x = var_951_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_57_cast_fp16")]; | |
| tensor<int32, [3]> var_956 = const()[name = tensor<string, []>("op_956"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_149_cast_fp16 = reshape(shape = var_956, x = attn_weights_57_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_151_cast_fp16 = softmax(axis = var_5, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")]; | |
| tensor<bool, []> attn_output_55_transpose_x_0 = const()[name = tensor<string, []>("attn_output_55_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_55_transpose_y_0 = const()[name = tensor<string, []>("attn_output_55_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_55_cast_fp16 = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151_cast_fp16, y = value_states_39_cast_fp16)[name = tensor<string, []>("attn_output_55_cast_fp16")]; | |
| tensor<int32, [4]> var_961 = const()[name = tensor<string, []>("op_961"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_57_cast_fp16 = reshape(shape = var_961, x = attn_output_55_cast_fp16)[name = tensor<string, []>("attn_output_57_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_59_perm_0 = const()[name = tensor<string, []>("attn_output_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_964 = const()[name = tensor<string, []>("op_964"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_8 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast_fp16)[name = tensor<string, []>("transpose_8")]; | |
| tensor<fp16, [1, 77, 768]> input_153_cast_fp16 = reshape(shape = var_964, x = transpose_8)[name = tensor<string, []>("input_153_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125218240))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125660672))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125660864)))]; | |
| tensor<fp16, [1, 77, 768]> linear_57_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_155_cast_fp16 = add(x = input_147_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")]; | |
| tensor<int32, [1]> input_157_axes_0 = const()[name = tensor<string, []>("input_157_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125662464)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125664064)))]; | |
| tensor<fp16, [1, 77, 768]> input_157_cast_fp16 = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_155_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125665664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127435200))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127435392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127437760))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_58_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")]; | |
| tensor<fp16, []> var_979_to_fp16 = const()[name = tensor<string, []>("op_979_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_980_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_979_to_fp16)[name = tensor<string, []>("op_980_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_981_cast_fp16 = sigmoid(x = var_980_cast_fp16)[name = tensor<string, []>("op_981_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_161_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_981_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127437952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129207488))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129207680)))]; | |
| tensor<fp16, [1, 77, 768]> linear_59_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_163_cast_fp16 = add(x = input_155_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129209280)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129210880)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_163_cast_fp16)[name = tensor<string, []>("hidden_states_61_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129212480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129654912))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129655104)))]; | |
| tensor<fp16, [1, 77, 768]> linear_60_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")]; | |
| tensor<fp16, []> var_1006_to_fp16 = const()[name = tensor<string, []>("op_1006_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_65_cast_fp16 = mul(x = linear_60_cast_fp16, y = var_1006_to_fp16)[name = tensor<string, []>("tensor_65_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129656704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130099136))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130099328)))]; | |
| tensor<fp16, [1, 77, 768]> linear_61_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1011 = const()[name = tensor<string, []>("op_1011"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1012_cast_fp16 = reshape(shape = var_1011, x = linear_61_cast_fp16)[name = tensor<string, []>("op_1012_cast_fp16")]; | |
| tensor<int32, [4]> var_1013_perm_0 = const()[name = tensor<string, []>("op_1013_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130100928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130543360))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130543552)))]; | |
| tensor<fp16, [1, 77, 768]> linear_62_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")]; | |
| tensor<int32, [4]> var_1018 = const()[name = tensor<string, []>("op_1018"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1019_cast_fp16 = reshape(shape = var_1018, x = linear_62_cast_fp16)[name = tensor<string, []>("op_1019_cast_fp16")]; | |
| tensor<int32, [4]> var_1020_perm_0 = const()[name = tensor<string, []>("op_1020_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1027 = const()[name = tensor<string, []>("op_1027"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1028_cast_fp16 = reshape(shape = var_1027, x = tensor_65_cast_fp16)[name = tensor<string, []>("op_1028_cast_fp16")]; | |
| tensor<int32, [4]> var_1029_perm_0 = const()[name = tensor<string, []>("op_1029_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1031 = const()[name = tensor<string, []>("op_1031"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_7 = transpose(perm = var_1029_perm_0, x = var_1028_cast_fp16)[name = tensor<string, []>("transpose_7")]; | |
| tensor<fp16, [12, 77, 64]> query_states_21_cast_fp16 = reshape(shape = var_1031, x = transpose_7)[name = tensor<string, []>("query_states_21_cast_fp16")]; | |
| tensor<int32, [3]> var_1033 = const()[name = tensor<string, []>("op_1033"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_6 = transpose(perm = var_1013_perm_0, x = var_1012_cast_fp16)[name = tensor<string, []>("transpose_6")]; | |
| tensor<fp16, [12, 77, 64]> key_states_43_cast_fp16 = reshape(shape = var_1033, x = transpose_6)[name = tensor<string, []>("key_states_43_cast_fp16")]; | |
| tensor<int32, [3]> var_1035 = const()[name = tensor<string, []>("op_1035"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_5 = transpose(perm = var_1020_perm_0, x = var_1019_cast_fp16)[name = tensor<string, []>("transpose_5")]; | |
| tensor<fp16, [12, 77, 64]> value_states_43_cast_fp16 = reshape(shape = var_1035, x = transpose_5)[name = tensor<string, []>("value_states_43_cast_fp16")]; | |
| tensor<bool, []> attn_weights_61_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_61_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_61_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_61_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_61_cast_fp16 = matmul(transpose_x = attn_weights_61_transpose_x_1, transpose_y = attn_weights_61_transpose_y_1, x = query_states_21_cast_fp16, y = key_states_43_cast_fp16)[name = tensor<string, []>("attn_weights_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1040 = const()[name = tensor<string, []>("op_1040"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1041_cast_fp16 = reshape(shape = var_1040, x = attn_weights_61_cast_fp16)[name = tensor<string, []>("op_1041_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_63_cast_fp16 = add(x = var_1041_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_63_cast_fp16")]; | |
| tensor<int32, [3]> var_1046 = const()[name = tensor<string, []>("op_1046"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_165_cast_fp16 = reshape(shape = var_1046, x = attn_weights_63_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_167_cast_fp16 = softmax(axis = var_5, x = input_165_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")]; | |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167_cast_fp16, y = value_states_43_cast_fp16)[name = tensor<string, []>("attn_output_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1051 = const()[name = tensor<string, []>("op_1051"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_63_cast_fp16 = reshape(shape = var_1051, x = attn_output_61_cast_fp16)[name = tensor<string, []>("attn_output_63_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_65_perm_0 = const()[name = tensor<string, []>("attn_output_65_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1054 = const()[name = tensor<string, []>("op_1054"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_4 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast_fp16)[name = tensor<string, []>("transpose_4")]; | |
| tensor<fp16, [1, 77, 768]> input_169_cast_fp16 = reshape(shape = var_1054, x = transpose_4)[name = tensor<string, []>("input_169_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130545152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130987584))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130987776)))]; | |
| tensor<fp16, [1, 77, 768]> linear_63_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")]; | |
| tensor<int32, [1]> input_173_axes_0 = const()[name = tensor<string, []>("input_173_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130989376)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130990976)))]; | |
| tensor<fp16, [1, 77, 768]> input_173_cast_fp16 = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_171_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130992576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132762112))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132762304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132764672))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_64_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")]; | |
| tensor<fp16, []> var_1069_to_fp16 = const()[name = tensor<string, []>("op_1069_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1070_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1069_to_fp16)[name = tensor<string, []>("op_1070_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_1071_cast_fp16 = sigmoid(x = var_1070_cast_fp16)[name = tensor<string, []>("op_1071_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_177_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1071_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132764864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134534400))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134534592)))]; | |
| tensor<fp16, [1, 77, 768]> linear_65_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_179_cast_fp16 = add(x = input_171_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")]; | |
| tensor<string, []> input_179_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("input_179_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = tensor<string, []>("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134536192)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134537792)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_179_cast_fp16)[name = tensor<string, []>("hidden_states_67_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134539392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134981824))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134982016)))]; | |
| tensor<fp16, [1, 77, 768]> linear_66_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")]; | |
| tensor<fp16, []> var_1096_to_fp16 = const()[name = tensor<string, []>("op_1096_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_cast_fp16 = mul(x = linear_66_cast_fp16, y = var_1096_to_fp16)[name = tensor<string, []>("tensor_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134983616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135426048))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135426240)))]; | |
| tensor<fp16, [1, 77, 768]> linear_67_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1101 = const()[name = tensor<string, []>("op_1101"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1102_cast_fp16 = reshape(shape = var_1101, x = linear_67_cast_fp16)[name = tensor<string, []>("op_1102_cast_fp16")]; | |
| tensor<int32, [4]> var_1103_perm_0 = const()[name = tensor<string, []>("op_1103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135427840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135870272))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135870464)))]; | |
| tensor<fp16, [1, 77, 768]> linear_68_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")]; | |
| tensor<int32, [4]> var_1108 = const()[name = tensor<string, []>("op_1108"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1109_cast_fp16 = reshape(shape = var_1108, x = linear_68_cast_fp16)[name = tensor<string, []>("op_1109_cast_fp16")]; | |
| tensor<int32, [4]> var_1110_perm_0 = const()[name = tensor<string, []>("op_1110_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1117 = const()[name = tensor<string, []>("op_1117"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1118_cast_fp16 = reshape(shape = var_1117, x = tensor_cast_fp16)[name = tensor<string, []>("op_1118_cast_fp16")]; | |
| tensor<int32, [4]> var_1119_perm_0 = const()[name = tensor<string, []>("op_1119_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1121 = const()[name = tensor<string, []>("op_1121"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_3 = transpose(perm = var_1119_perm_0, x = var_1118_cast_fp16)[name = tensor<string, []>("transpose_3")]; | |
| tensor<fp16, [12, 77, 64]> query_states_cast_fp16 = reshape(shape = var_1121, x = transpose_3)[name = tensor<string, []>("query_states_cast_fp16")]; | |
| tensor<int32, [3]> var_1123 = const()[name = tensor<string, []>("op_1123"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_2 = transpose(perm = var_1103_perm_0, x = var_1102_cast_fp16)[name = tensor<string, []>("transpose_2")]; | |
| tensor<fp16, [12, 77, 64]> key_states_cast_fp16 = reshape(shape = var_1123, x = transpose_2)[name = tensor<string, []>("key_states_cast_fp16")]; | |
| tensor<int32, [3]> var_1125 = const()[name = tensor<string, []>("op_1125"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_1 = transpose(perm = var_1110_perm_0, x = var_1109_cast_fp16)[name = tensor<string, []>("transpose_1")]; | |
| tensor<fp16, [12, 77, 64]> value_states_cast_fp16 = reshape(shape = var_1125, x = transpose_1)[name = tensor<string, []>("value_states_cast_fp16")]; | |
| tensor<bool, []> attn_weights_67_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_67_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_67_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_67_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_67_cast_fp16 = matmul(transpose_x = attn_weights_67_transpose_x_1, transpose_y = attn_weights_67_transpose_y_1, x = query_states_cast_fp16, y = key_states_cast_fp16)[name = tensor<string, []>("attn_weights_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1130 = const()[name = tensor<string, []>("op_1130"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1131_cast_fp16 = reshape(shape = var_1130, x = attn_weights_67_cast_fp16)[name = tensor<string, []>("op_1131_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_69_cast_fp16 = add(x = var_1131_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_69_cast_fp16")]; | |
| tensor<int32, [3]> var_1136 = const()[name = tensor<string, []>("op_1136"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_181_cast_fp16 = reshape(shape = var_1136, x = attn_weights_69_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_183_cast_fp16 = softmax(axis = var_5, x = input_181_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")]; | |
| tensor<bool, []> attn_output_67_transpose_x_0 = const()[name = tensor<string, []>("attn_output_67_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_67_transpose_y_0 = const()[name = tensor<string, []>("attn_output_67_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_67_cast_fp16 = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183_cast_fp16, y = value_states_cast_fp16)[name = tensor<string, []>("attn_output_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1141 = const()[name = tensor<string, []>("op_1141"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_69_cast_fp16 = reshape(shape = var_1141, x = attn_output_67_cast_fp16)[name = tensor<string, []>("attn_output_69_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_perm_0 = const()[name = tensor<string, []>("attn_output_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1144 = const()[name = tensor<string, []>("op_1144"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_0 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast_fp16)[name = tensor<string, []>("transpose_0")]; | |
| tensor<fp16, [1, 77, 768]> input_185_cast_fp16 = reshape(shape = var_1144, x = transpose_0)[name = tensor<string, []>("input_185_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135872064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136314496))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136314688)))]; | |
| tensor<fp16, [1, 77, 768]> linear_69_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_187_cast_fp16 = add(x = input_179_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")]; | |
| tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136316288)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136317888)))]; | |
| tensor<fp16, [1, 77, 768]> input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136319488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138089024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138089216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138091584))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> linear_70_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")]; | |
| tensor<fp16, []> var_1159_to_fp16 = const()[name = tensor<string, []>("op_1159_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1160_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1159_to_fp16)[name = tensor<string, []>("op_1160_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_1161_cast_fp16 = sigmoid(x = var_1160_cast_fp16)[name = tensor<string, []>("op_1161_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_193_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1161_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138091776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139861312))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139861504)))]; | |
| tensor<fp16, [1, 77, 768]> linear_71_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_cast_fp16 = add(x = input_187_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; | |
| tensor<int32, [1]> last_hidden_state_axes_0 = const()[name = tensor<string, []>("last_hidden_state_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139863104)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139864704)))]; | |
| tensor<fp16, [1, 77, 768]> last_hidden_state_cast_fp16 = layer_norm(axes = last_hidden_state_axes_0, beta = text_encoder_text_model_final_layer_norm_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_final_layer_norm_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("last_hidden_state_cast_fp16")]; | |
| tensor<int32, [1]> var_1175 = const()[name = tensor<string, []>("op_1175"), val = tensor<int32, [1]>([0])]; | |
| tensor<int32, [1]> var_1177 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_2)[name = tensor<string, []>("op_1177")]; | |
| tensor<int32, []> stack_0_axis_0 = const()[name = tensor<string, []>("stack_0_axis_0"), val = tensor<int32, []>(1)]; | |
| tensor<int32, [1, 2]> stack_0 = stack(axis = stack_0_axis_0, values = (var_1175, var_1177))[name = tensor<string, []>("stack_0")]; | |
| tensor<int32, []> var_1179_transpose_batch_dims_0 = const()[name = tensor<string, []>("op_1179_transpose_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [1, 768]> var_1179_transpose_cast_fp16 = gather_nd(batch_dims = var_1179_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor<string, []>("op_1179_transpose_cast_fp16")]; | |
| tensor<string, []> var_1179_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_1179_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<fp32, [1, 77, 768]> hidden_embeds = cast(dtype = input_179_cast_fp16_to_fp32_dtype_0, x = input_179_cast_fp16)[name = tensor<string, []>("cast_0")]; | |
| tensor<fp32, [1, 768]> pooled_outputs = cast(dtype = var_1179_cast_fp16_to_fp32_dtype_0, x = var_1179_transpose_cast_fp16)[name = tensor<string, []>("cast_1")]; | |
| } -> (hidden_embeds, pooled_outputs); | |
| } |