diff --git "a/ggml-large-v3-turbo-encoder.mlmodelc/model1/model.mil" "b/ggml-large-v3-turbo-encoder.mlmodelc/model1/model.mil" new file mode 100644--- /dev/null +++ "b/ggml-large-v3-turbo-encoder.mlmodelc/model1/model.mil" @@ -0,0 +1,769 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3400.43.1"}, {"coremlc-version", "3400.58.2"}})] +{ + func main(tensor linear_95_cast_fp16, tensor x_193_cast_fp16) { + tensor linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(64)))]; + string cast_0_dtype_0 = const()[name = string("cast_0_dtype_0"), val = string("fp16")]; + string cast_1_dtype_0 = const()[name = string("cast_1_dtype_0"), val = string("fp16")]; + tensor cast_1 = cast(dtype = cast_1_dtype_0, x = linear_95_cast_fp16)[name = string("cast_0")]; + tensor cast_0 = cast(dtype = cast_0_dtype_0, x = x_193_cast_fp16)[name = string("cast_1")]; + tensor x_199_cast_fp16 = add(x = cast_0, y = cast_1)[name = string("x_199_cast_fp16")]; + tensor var_1659_axes_0 = const()[name = string("op_1659_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = string("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(2688)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = string("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(5312)))]; + fp16 var_1649_to_fp16 = const()[name = string("op_1649_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1659_cast_fp16 = layer_norm(axes = var_1659_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1649_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = string("op_1659_cast_fp16")]; + tensor var_1670_to_fp16 = const()[name = string("op_1670_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(7936)))]; + tensor var_1671_to_fp16 = const()[name = string("op_1671_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(3284800)))]; + tensor linear_96_cast_fp16 = linear(bias = var_1671_to_fp16, weight = var_1670_to_fp16, x = var_1659_cast_fp16)[name = string("linear_96_cast_fp16")]; + tensor var_1674_to_fp16 = const()[name = string("op_1674_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(3287424)))]; + tensor linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1674_to_fp16, x = var_1659_cast_fp16)[name = string("linear_97_cast_fp16")]; + tensor var_1678_to_fp16 = const()[name = string("op_1678_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(6564288)))]; + tensor var_1679_to_fp16 = const()[name = string("op_1679_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(9841152)))]; + tensor linear_98_cast_fp16 = linear(bias = var_1679_to_fp16, weight = var_1678_to_fp16, x = var_1659_cast_fp16)[name = string("linear_98_cast_fp16")]; + tensor var_1683 = const()[name = string("op_1683"), val = tensor([1, 1500, 20, -1])]; + tensor var_1684_cast_fp16 = reshape(shape = var_1683, x = linear_96_cast_fp16)[name = string("op_1684_cast_fp16")]; + tensor var_1689 = const()[name = string("op_1689"), val = tensor([1, 1500, 20, -1])]; + tensor var_1690_cast_fp16 = reshape(shape = var_1689, x = linear_97_cast_fp16)[name = string("op_1690_cast_fp16")]; + tensor var_1695 = const()[name = string("op_1695"), val = tensor([1, 1500, 20, -1])]; + tensor var_1696_cast_fp16 = reshape(shape = var_1695, x = linear_98_cast_fp16)[name = string("op_1696_cast_fp16")]; + tensor transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_98 = transpose(perm = transpose_98_perm_0, x = var_1696_cast_fp16)[name = string("transpose_205")]; + tensor transpose_97 = transpose(perm = transpose_97_perm_0, x = var_1690_cast_fp16)[name = string("transpose_206")]; + tensor transpose_96 = transpose(perm = transpose_96_perm_0, x = var_1684_cast_fp16)[name = string("transpose_207")]; + tensor a_33_cast_fp16 = scaled_dot_product_attention(key = transpose_97, query = transpose_96, value = transpose_98)[name = string("a_33_cast_fp16")]; + tensor var_1700 = const()[name = string("op_1700"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = string("concat_16"), val = tensor([1, 1500, 1280])]; + tensor var_1701_cast_fp16 = transpose(perm = var_1700, x = a_33_cast_fp16)[name = string("transpose_204")]; + tensor x_203_cast_fp16 = reshape(shape = concat_16, x = var_1701_cast_fp16)[name = string("x_203_cast_fp16")]; + tensor var_1705_to_fp16 = const()[name = string("op_1705_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(9843776)))]; + tensor var_1706_to_fp16 = const()[name = string("op_1706_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(13120640)))]; + tensor linear_99_cast_fp16 = linear(bias = var_1706_to_fp16, weight = var_1705_to_fp16, x = x_203_cast_fp16)[name = string("linear_99_cast_fp16")]; + tensor x_205_cast_fp16 = add(x = x_199_cast_fp16, y = linear_99_cast_fp16)[name = string("x_205_cast_fp16")]; + tensor var_1713_axes_0 = const()[name = string("op_1713_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = string("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(13123264)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = string("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(13125888)))]; + tensor var_1713_cast_fp16 = layer_norm(axes = var_1713_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1649_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = string("op_1713_cast_fp16")]; + tensor var_1722_to_fp16 = const()[name = string("op_1722_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(13128512)))]; + tensor var_1723_to_fp16 = const()[name = string("op_1723_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(26235776)))]; + tensor linear_100_cast_fp16 = linear(bias = var_1723_to_fp16, weight = var_1722_to_fp16, x = var_1713_cast_fp16)[name = string("linear_100_cast_fp16")]; + string x_209_mode_0 = const()[name = string("x_209_mode_0"), val = string("EXACT")]; + tensor x_209_cast_fp16 = gelu(mode = x_209_mode_0, x = linear_100_cast_fp16)[name = string("x_209_cast_fp16")]; + tensor var_1728_to_fp16 = const()[name = string("op_1728_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(26246080)))]; + tensor var_1729_to_fp16 = const()[name = string("op_1729_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(39353344)))]; + tensor linear_101_cast_fp16 = linear(bias = var_1729_to_fp16, weight = var_1728_to_fp16, x = x_209_cast_fp16)[name = string("linear_101_cast_fp16")]; + tensor x_211_cast_fp16 = add(x = x_205_cast_fp16, y = linear_101_cast_fp16)[name = string("x_211_cast_fp16")]; + tensor var_1754_axes_0 = const()[name = string("op_1754_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = string("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(39355968)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = string("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(39358592)))]; + fp16 var_1744_to_fp16 = const()[name = string("op_1744_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1754_cast_fp16 = layer_norm(axes = var_1754_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1744_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = string("op_1754_cast_fp16")]; + tensor var_1765_to_fp16 = const()[name = string("op_1765_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(39361216)))]; + tensor var_1766_to_fp16 = const()[name = string("op_1766_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(42638080)))]; + tensor linear_102_cast_fp16 = linear(bias = var_1766_to_fp16, weight = var_1765_to_fp16, x = var_1754_cast_fp16)[name = string("linear_102_cast_fp16")]; + tensor var_1769_to_fp16 = const()[name = string("op_1769_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(42640704)))]; + tensor linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1769_to_fp16, x = var_1754_cast_fp16)[name = string("linear_103_cast_fp16")]; + tensor var_1773_to_fp16 = const()[name = string("op_1773_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(45917568)))]; + tensor var_1774_to_fp16 = const()[name = string("op_1774_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(49194432)))]; + tensor linear_104_cast_fp16 = linear(bias = var_1774_to_fp16, weight = var_1773_to_fp16, x = var_1754_cast_fp16)[name = string("linear_104_cast_fp16")]; + tensor var_1778 = const()[name = string("op_1778"), val = tensor([1, 1500, 20, -1])]; + tensor var_1779_cast_fp16 = reshape(shape = var_1778, x = linear_102_cast_fp16)[name = string("op_1779_cast_fp16")]; + tensor var_1784 = const()[name = string("op_1784"), val = tensor([1, 1500, 20, -1])]; + tensor var_1785_cast_fp16 = reshape(shape = var_1784, x = linear_103_cast_fp16)[name = string("op_1785_cast_fp16")]; + tensor var_1790 = const()[name = string("op_1790"), val = tensor([1, 1500, 20, -1])]; + tensor var_1791_cast_fp16 = reshape(shape = var_1790, x = linear_104_cast_fp16)[name = string("op_1791_cast_fp16")]; + tensor transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_100_perm_0 = const()[name = string("transpose_100_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_101_perm_0 = const()[name = string("transpose_101_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_101 = transpose(perm = transpose_101_perm_0, x = var_1791_cast_fp16)[name = string("transpose_201")]; + tensor transpose_100 = transpose(perm = transpose_100_perm_0, x = var_1785_cast_fp16)[name = string("transpose_202")]; + tensor transpose_99 = transpose(perm = transpose_99_perm_0, x = var_1779_cast_fp16)[name = string("transpose_203")]; + tensor a_35_cast_fp16 = scaled_dot_product_attention(key = transpose_100, query = transpose_99, value = transpose_101)[name = string("a_35_cast_fp16")]; + tensor var_1795 = const()[name = string("op_1795"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = string("concat_17"), val = tensor([1, 1500, 1280])]; + tensor var_1796_cast_fp16 = transpose(perm = var_1795, x = a_35_cast_fp16)[name = string("transpose_200")]; + tensor x_215_cast_fp16 = reshape(shape = concat_17, x = var_1796_cast_fp16)[name = string("x_215_cast_fp16")]; + tensor var_1800_to_fp16 = const()[name = string("op_1800_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(49197056)))]; + tensor var_1801_to_fp16 = const()[name = string("op_1801_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(52473920)))]; + tensor linear_105_cast_fp16 = linear(bias = var_1801_to_fp16, weight = var_1800_to_fp16, x = x_215_cast_fp16)[name = string("linear_105_cast_fp16")]; + tensor x_217_cast_fp16 = add(x = x_211_cast_fp16, y = linear_105_cast_fp16)[name = string("x_217_cast_fp16")]; + tensor var_1808_axes_0 = const()[name = string("op_1808_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = string("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(52476544)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = string("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(52479168)))]; + tensor var_1808_cast_fp16 = layer_norm(axes = var_1808_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1744_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = string("op_1808_cast_fp16")]; + tensor var_1817_to_fp16 = const()[name = string("op_1817_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(52481792)))]; + tensor var_1818_to_fp16 = const()[name = string("op_1818_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(65589056)))]; + tensor linear_106_cast_fp16 = linear(bias = var_1818_to_fp16, weight = var_1817_to_fp16, x = var_1808_cast_fp16)[name = string("linear_106_cast_fp16")]; + string x_221_mode_0 = const()[name = string("x_221_mode_0"), val = string("EXACT")]; + tensor x_221_cast_fp16 = gelu(mode = x_221_mode_0, x = linear_106_cast_fp16)[name = string("x_221_cast_fp16")]; + tensor var_1823_to_fp16 = const()[name = string("op_1823_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(65599360)))]; + tensor var_1824_to_fp16 = const()[name = string("op_1824_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(78706624)))]; + tensor linear_107_cast_fp16 = linear(bias = var_1824_to_fp16, weight = var_1823_to_fp16, x = x_221_cast_fp16)[name = string("linear_107_cast_fp16")]; + tensor x_223_cast_fp16 = add(x = x_217_cast_fp16, y = linear_107_cast_fp16)[name = string("x_223_cast_fp16")]; + tensor var_1849_axes_0 = const()[name = string("op_1849_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = string("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(78709248)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = string("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(78711872)))]; + fp16 var_1839_to_fp16 = const()[name = string("op_1839_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1849_cast_fp16 = layer_norm(axes = var_1849_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_1839_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = string("op_1849_cast_fp16")]; + tensor var_1860_to_fp16 = const()[name = string("op_1860_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(78714496)))]; + tensor var_1861_to_fp16 = const()[name = string("op_1861_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(81991360)))]; + tensor linear_108_cast_fp16 = linear(bias = var_1861_to_fp16, weight = var_1860_to_fp16, x = var_1849_cast_fp16)[name = string("linear_108_cast_fp16")]; + tensor var_1864_to_fp16 = const()[name = string("op_1864_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(81993984)))]; + tensor linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1864_to_fp16, x = var_1849_cast_fp16)[name = string("linear_109_cast_fp16")]; + tensor var_1868_to_fp16 = const()[name = string("op_1868_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(85270848)))]; + tensor var_1869_to_fp16 = const()[name = string("op_1869_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(88547712)))]; + tensor linear_110_cast_fp16 = linear(bias = var_1869_to_fp16, weight = var_1868_to_fp16, x = var_1849_cast_fp16)[name = string("linear_110_cast_fp16")]; + tensor var_1873 = const()[name = string("op_1873"), val = tensor([1, 1500, 20, -1])]; + tensor var_1874_cast_fp16 = reshape(shape = var_1873, x = linear_108_cast_fp16)[name = string("op_1874_cast_fp16")]; + tensor var_1879 = const()[name = string("op_1879"), val = tensor([1, 1500, 20, -1])]; + tensor var_1880_cast_fp16 = reshape(shape = var_1879, x = linear_109_cast_fp16)[name = string("op_1880_cast_fp16")]; + tensor var_1885 = const()[name = string("op_1885"), val = tensor([1, 1500, 20, -1])]; + tensor var_1886_cast_fp16 = reshape(shape = var_1885, x = linear_110_cast_fp16)[name = string("op_1886_cast_fp16")]; + tensor transpose_102_perm_0 = const()[name = string("transpose_102_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_103_perm_0 = const()[name = string("transpose_103_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_104_perm_0 = const()[name = string("transpose_104_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_104 = transpose(perm = transpose_104_perm_0, x = var_1886_cast_fp16)[name = string("transpose_197")]; + tensor transpose_103 = transpose(perm = transpose_103_perm_0, x = var_1880_cast_fp16)[name = string("transpose_198")]; + tensor transpose_102 = transpose(perm = transpose_102_perm_0, x = var_1874_cast_fp16)[name = string("transpose_199")]; + tensor a_37_cast_fp16 = scaled_dot_product_attention(key = transpose_103, query = transpose_102, value = transpose_104)[name = string("a_37_cast_fp16")]; + tensor var_1890 = const()[name = string("op_1890"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = string("concat_18"), val = tensor([1, 1500, 1280])]; + tensor var_1891_cast_fp16 = transpose(perm = var_1890, x = a_37_cast_fp16)[name = string("transpose_196")]; + tensor x_227_cast_fp16 = reshape(shape = concat_18, x = var_1891_cast_fp16)[name = string("x_227_cast_fp16")]; + tensor var_1895_to_fp16 = const()[name = string("op_1895_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(88550336)))]; + tensor var_1896_to_fp16 = const()[name = string("op_1896_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(91827200)))]; + tensor linear_111_cast_fp16 = linear(bias = var_1896_to_fp16, weight = var_1895_to_fp16, x = x_227_cast_fp16)[name = string("linear_111_cast_fp16")]; + tensor x_229_cast_fp16 = add(x = x_223_cast_fp16, y = linear_111_cast_fp16)[name = string("x_229_cast_fp16")]; + tensor var_1903_axes_0 = const()[name = string("op_1903_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = string("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(91829824)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = string("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(91832448)))]; + tensor var_1903_cast_fp16 = layer_norm(axes = var_1903_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_1839_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = string("op_1903_cast_fp16")]; + tensor var_1912_to_fp16 = const()[name = string("op_1912_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(91835072)))]; + tensor var_1913_to_fp16 = const()[name = string("op_1913_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(104942336)))]; + tensor linear_112_cast_fp16 = linear(bias = var_1913_to_fp16, weight = var_1912_to_fp16, x = var_1903_cast_fp16)[name = string("linear_112_cast_fp16")]; + string x_233_mode_0 = const()[name = string("x_233_mode_0"), val = string("EXACT")]; + tensor x_233_cast_fp16 = gelu(mode = x_233_mode_0, x = linear_112_cast_fp16)[name = string("x_233_cast_fp16")]; + tensor var_1918_to_fp16 = const()[name = string("op_1918_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(104952640)))]; + tensor var_1919_to_fp16 = const()[name = string("op_1919_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(118059904)))]; + tensor linear_113_cast_fp16 = linear(bias = var_1919_to_fp16, weight = var_1918_to_fp16, x = x_233_cast_fp16)[name = string("linear_113_cast_fp16")]; + tensor x_235_cast_fp16 = add(x = x_229_cast_fp16, y = linear_113_cast_fp16)[name = string("x_235_cast_fp16")]; + tensor var_1944_axes_0 = const()[name = string("op_1944_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = string("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(118062528)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = string("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(118065152)))]; + fp16 var_1934_to_fp16 = const()[name = string("op_1934_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1944_cast_fp16 = layer_norm(axes = var_1944_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_1934_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = string("op_1944_cast_fp16")]; + tensor var_1955_to_fp16 = const()[name = string("op_1955_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(118067776)))]; + tensor var_1956_to_fp16 = const()[name = string("op_1956_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(121344640)))]; + tensor linear_114_cast_fp16 = linear(bias = var_1956_to_fp16, weight = var_1955_to_fp16, x = var_1944_cast_fp16)[name = string("linear_114_cast_fp16")]; + tensor var_1959_to_fp16 = const()[name = string("op_1959_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(121347264)))]; + tensor linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1959_to_fp16, x = var_1944_cast_fp16)[name = string("linear_115_cast_fp16")]; + tensor var_1963_to_fp16 = const()[name = string("op_1963_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(124624128)))]; + tensor var_1964_to_fp16 = const()[name = string("op_1964_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(127900992)))]; + tensor linear_116_cast_fp16 = linear(bias = var_1964_to_fp16, weight = var_1963_to_fp16, x = var_1944_cast_fp16)[name = string("linear_116_cast_fp16")]; + tensor var_1968 = const()[name = string("op_1968"), val = tensor([1, 1500, 20, -1])]; + tensor var_1969_cast_fp16 = reshape(shape = var_1968, x = linear_114_cast_fp16)[name = string("op_1969_cast_fp16")]; + tensor var_1974 = const()[name = string("op_1974"), val = tensor([1, 1500, 20, -1])]; + tensor var_1975_cast_fp16 = reshape(shape = var_1974, x = linear_115_cast_fp16)[name = string("op_1975_cast_fp16")]; + tensor var_1980 = const()[name = string("op_1980"), val = tensor([1, 1500, 20, -1])]; + tensor var_1981_cast_fp16 = reshape(shape = var_1980, x = linear_116_cast_fp16)[name = string("op_1981_cast_fp16")]; + tensor transpose_105_perm_0 = const()[name = string("transpose_105_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_106_perm_0 = const()[name = string("transpose_106_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_107_perm_0 = const()[name = string("transpose_107_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_107 = transpose(perm = transpose_107_perm_0, x = var_1981_cast_fp16)[name = string("transpose_193")]; + tensor transpose_106 = transpose(perm = transpose_106_perm_0, x = var_1975_cast_fp16)[name = string("transpose_194")]; + tensor transpose_105 = transpose(perm = transpose_105_perm_0, x = var_1969_cast_fp16)[name = string("transpose_195")]; + tensor a_39_cast_fp16 = scaled_dot_product_attention(key = transpose_106, query = transpose_105, value = transpose_107)[name = string("a_39_cast_fp16")]; + tensor var_1985 = const()[name = string("op_1985"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = string("concat_19"), val = tensor([1, 1500, 1280])]; + tensor var_1986_cast_fp16 = transpose(perm = var_1985, x = a_39_cast_fp16)[name = string("transpose_192")]; + tensor x_239_cast_fp16 = reshape(shape = concat_19, x = var_1986_cast_fp16)[name = string("x_239_cast_fp16")]; + tensor var_1990_to_fp16 = const()[name = string("op_1990_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(127903616)))]; + tensor var_1991_to_fp16 = const()[name = string("op_1991_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(131180480)))]; + tensor linear_117_cast_fp16 = linear(bias = var_1991_to_fp16, weight = var_1990_to_fp16, x = x_239_cast_fp16)[name = string("linear_117_cast_fp16")]; + tensor x_241_cast_fp16 = add(x = x_235_cast_fp16, y = linear_117_cast_fp16)[name = string("x_241_cast_fp16")]; + tensor var_1998_axes_0 = const()[name = string("op_1998_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = string("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(131183104)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = string("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(131185728)))]; + tensor var_1998_cast_fp16 = layer_norm(axes = var_1998_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_1934_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = string("op_1998_cast_fp16")]; + tensor var_2007_to_fp16 = const()[name = string("op_2007_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(131188352)))]; + tensor var_2008_to_fp16 = const()[name = string("op_2008_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(144295616)))]; + tensor linear_118_cast_fp16 = linear(bias = var_2008_to_fp16, weight = var_2007_to_fp16, x = var_1998_cast_fp16)[name = string("linear_118_cast_fp16")]; + string x_245_mode_0 = const()[name = string("x_245_mode_0"), val = string("EXACT")]; + tensor x_245_cast_fp16 = gelu(mode = x_245_mode_0, x = linear_118_cast_fp16)[name = string("x_245_cast_fp16")]; + tensor var_2013_to_fp16 = const()[name = string("op_2013_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(144305920)))]; + tensor var_2014_to_fp16 = const()[name = string("op_2014_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(157413184)))]; + tensor linear_119_cast_fp16 = linear(bias = var_2014_to_fp16, weight = var_2013_to_fp16, x = x_245_cast_fp16)[name = string("linear_119_cast_fp16")]; + tensor x_247_cast_fp16 = add(x = x_241_cast_fp16, y = linear_119_cast_fp16)[name = string("x_247_cast_fp16")]; + tensor var_2039_axes_0 = const()[name = string("op_2039_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = string("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(157415808)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = string("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(157418432)))]; + fp16 var_2029_to_fp16 = const()[name = string("op_2029_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2039_cast_fp16 = layer_norm(axes = var_2039_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2029_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = string("op_2039_cast_fp16")]; + tensor var_2050_to_fp16 = const()[name = string("op_2050_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(157421056)))]; + tensor var_2051_to_fp16 = const()[name = string("op_2051_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(160697920)))]; + tensor linear_120_cast_fp16 = linear(bias = var_2051_to_fp16, weight = var_2050_to_fp16, x = var_2039_cast_fp16)[name = string("linear_120_cast_fp16")]; + tensor var_2054_to_fp16 = const()[name = string("op_2054_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(160700544)))]; + tensor linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2054_to_fp16, x = var_2039_cast_fp16)[name = string("linear_121_cast_fp16")]; + tensor var_2058_to_fp16 = const()[name = string("op_2058_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(163977408)))]; + tensor var_2059_to_fp16 = const()[name = string("op_2059_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(167254272)))]; + tensor linear_122_cast_fp16 = linear(bias = var_2059_to_fp16, weight = var_2058_to_fp16, x = var_2039_cast_fp16)[name = string("linear_122_cast_fp16")]; + tensor var_2063 = const()[name = string("op_2063"), val = tensor([1, 1500, 20, -1])]; + tensor var_2064_cast_fp16 = reshape(shape = var_2063, x = linear_120_cast_fp16)[name = string("op_2064_cast_fp16")]; + tensor var_2069 = const()[name = string("op_2069"), val = tensor([1, 1500, 20, -1])]; + tensor var_2070_cast_fp16 = reshape(shape = var_2069, x = linear_121_cast_fp16)[name = string("op_2070_cast_fp16")]; + tensor var_2075 = const()[name = string("op_2075"), val = tensor([1, 1500, 20, -1])]; + tensor var_2076_cast_fp16 = reshape(shape = var_2075, x = linear_122_cast_fp16)[name = string("op_2076_cast_fp16")]; + tensor transpose_108_perm_0 = const()[name = string("transpose_108_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_109_perm_0 = const()[name = string("transpose_109_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_110_perm_0 = const()[name = string("transpose_110_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_110 = transpose(perm = transpose_110_perm_0, x = var_2076_cast_fp16)[name = string("transpose_189")]; + tensor transpose_109 = transpose(perm = transpose_109_perm_0, x = var_2070_cast_fp16)[name = string("transpose_190")]; + tensor transpose_108 = transpose(perm = transpose_108_perm_0, x = var_2064_cast_fp16)[name = string("transpose_191")]; + tensor a_41_cast_fp16 = scaled_dot_product_attention(key = transpose_109, query = transpose_108, value = transpose_110)[name = string("a_41_cast_fp16")]; + tensor var_2080 = const()[name = string("op_2080"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = string("concat_20"), val = tensor([1, 1500, 1280])]; + tensor var_2081_cast_fp16 = transpose(perm = var_2080, x = a_41_cast_fp16)[name = string("transpose_188")]; + tensor x_251_cast_fp16 = reshape(shape = concat_20, x = var_2081_cast_fp16)[name = string("x_251_cast_fp16")]; + tensor var_2085_to_fp16 = const()[name = string("op_2085_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(167256896)))]; + tensor var_2086_to_fp16 = const()[name = string("op_2086_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(170533760)))]; + tensor linear_123_cast_fp16 = linear(bias = var_2086_to_fp16, weight = var_2085_to_fp16, x = x_251_cast_fp16)[name = string("linear_123_cast_fp16")]; + tensor x_253_cast_fp16 = add(x = x_247_cast_fp16, y = linear_123_cast_fp16)[name = string("x_253_cast_fp16")]; + tensor var_2093_axes_0 = const()[name = string("op_2093_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = string("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(170536384)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = string("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(170539008)))]; + tensor var_2093_cast_fp16 = layer_norm(axes = var_2093_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2029_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = string("op_2093_cast_fp16")]; + tensor var_2102_to_fp16 = const()[name = string("op_2102_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(170541632)))]; + tensor var_2103_to_fp16 = const()[name = string("op_2103_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(183648896)))]; + tensor linear_124_cast_fp16 = linear(bias = var_2103_to_fp16, weight = var_2102_to_fp16, x = var_2093_cast_fp16)[name = string("linear_124_cast_fp16")]; + string x_257_mode_0 = const()[name = string("x_257_mode_0"), val = string("EXACT")]; + tensor x_257_cast_fp16 = gelu(mode = x_257_mode_0, x = linear_124_cast_fp16)[name = string("x_257_cast_fp16")]; + tensor var_2108_to_fp16 = const()[name = string("op_2108_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(183659200)))]; + tensor var_2109_to_fp16 = const()[name = string("op_2109_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(196766464)))]; + tensor linear_125_cast_fp16 = linear(bias = var_2109_to_fp16, weight = var_2108_to_fp16, x = x_257_cast_fp16)[name = string("linear_125_cast_fp16")]; + tensor x_259_cast_fp16 = add(x = x_253_cast_fp16, y = linear_125_cast_fp16)[name = string("x_259_cast_fp16")]; + tensor var_2134_axes_0 = const()[name = string("op_2134_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = string("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(196769088)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = string("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(196771712)))]; + fp16 var_2124_to_fp16 = const()[name = string("op_2124_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2134_cast_fp16 = layer_norm(axes = var_2134_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2124_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = string("op_2134_cast_fp16")]; + tensor var_2145_to_fp16 = const()[name = string("op_2145_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(196774336)))]; + tensor var_2146_to_fp16 = const()[name = string("op_2146_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(200051200)))]; + tensor linear_126_cast_fp16 = linear(bias = var_2146_to_fp16, weight = var_2145_to_fp16, x = var_2134_cast_fp16)[name = string("linear_126_cast_fp16")]; + tensor var_2149_to_fp16 = const()[name = string("op_2149_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(200053824)))]; + tensor linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2149_to_fp16, x = var_2134_cast_fp16)[name = string("linear_127_cast_fp16")]; + tensor var_2153_to_fp16 = const()[name = string("op_2153_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(203330688)))]; + tensor var_2154_to_fp16 = const()[name = string("op_2154_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(206607552)))]; + tensor linear_128_cast_fp16 = linear(bias = var_2154_to_fp16, weight = var_2153_to_fp16, x = var_2134_cast_fp16)[name = string("linear_128_cast_fp16")]; + tensor var_2158 = const()[name = string("op_2158"), val = tensor([1, 1500, 20, -1])]; + tensor var_2159_cast_fp16 = reshape(shape = var_2158, x = linear_126_cast_fp16)[name = string("op_2159_cast_fp16")]; + tensor var_2164 = const()[name = string("op_2164"), val = tensor([1, 1500, 20, -1])]; + tensor var_2165_cast_fp16 = reshape(shape = var_2164, x = linear_127_cast_fp16)[name = string("op_2165_cast_fp16")]; + tensor var_2170 = const()[name = string("op_2170"), val = tensor([1, 1500, 20, -1])]; + tensor var_2171_cast_fp16 = reshape(shape = var_2170, x = linear_128_cast_fp16)[name = string("op_2171_cast_fp16")]; + tensor transpose_111_perm_0 = const()[name = string("transpose_111_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_112_perm_0 = const()[name = string("transpose_112_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_113_perm_0 = const()[name = string("transpose_113_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_113 = transpose(perm = transpose_113_perm_0, x = var_2171_cast_fp16)[name = string("transpose_185")]; + tensor transpose_112 = transpose(perm = transpose_112_perm_0, x = var_2165_cast_fp16)[name = string("transpose_186")]; + tensor transpose_111 = transpose(perm = transpose_111_perm_0, x = var_2159_cast_fp16)[name = string("transpose_187")]; + tensor a_43_cast_fp16 = scaled_dot_product_attention(key = transpose_112, query = transpose_111, value = transpose_113)[name = string("a_43_cast_fp16")]; + tensor var_2175 = const()[name = string("op_2175"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = string("concat_21"), val = tensor([1, 1500, 1280])]; + tensor var_2176_cast_fp16 = transpose(perm = var_2175, x = a_43_cast_fp16)[name = string("transpose_184")]; + tensor x_263_cast_fp16 = reshape(shape = concat_21, x = var_2176_cast_fp16)[name = string("x_263_cast_fp16")]; + tensor var_2180_to_fp16 = const()[name = string("op_2180_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(206610176)))]; + tensor var_2181_to_fp16 = const()[name = string("op_2181_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(209887040)))]; + tensor linear_129_cast_fp16 = linear(bias = var_2181_to_fp16, weight = var_2180_to_fp16, x = x_263_cast_fp16)[name = string("linear_129_cast_fp16")]; + tensor x_265_cast_fp16 = add(x = x_259_cast_fp16, y = linear_129_cast_fp16)[name = string("x_265_cast_fp16")]; + tensor var_2188_axes_0 = const()[name = string("op_2188_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = string("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(209889664)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = string("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(209892288)))]; + tensor var_2188_cast_fp16 = layer_norm(axes = var_2188_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2124_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = string("op_2188_cast_fp16")]; + tensor var_2197_to_fp16 = const()[name = string("op_2197_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(209894912)))]; + tensor var_2198_to_fp16 = const()[name = string("op_2198_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(223002176)))]; + tensor linear_130_cast_fp16 = linear(bias = var_2198_to_fp16, weight = var_2197_to_fp16, x = var_2188_cast_fp16)[name = string("linear_130_cast_fp16")]; + string x_269_mode_0 = const()[name = string("x_269_mode_0"), val = string("EXACT")]; + tensor x_269_cast_fp16 = gelu(mode = x_269_mode_0, x = linear_130_cast_fp16)[name = string("x_269_cast_fp16")]; + tensor var_2203_to_fp16 = const()[name = string("op_2203_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(223012480)))]; + tensor var_2204_to_fp16 = const()[name = string("op_2204_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(236119744)))]; + tensor linear_131_cast_fp16 = linear(bias = var_2204_to_fp16, weight = var_2203_to_fp16, x = x_269_cast_fp16)[name = string("linear_131_cast_fp16")]; + tensor x_271_cast_fp16 = add(x = x_265_cast_fp16, y = linear_131_cast_fp16)[name = string("x_271_cast_fp16")]; + tensor var_2229_axes_0 = const()[name = string("op_2229_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = string("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(236122368)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = string("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(236124992)))]; + fp16 var_2219_to_fp16 = const()[name = string("op_2219_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2229_cast_fp16 = layer_norm(axes = var_2229_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2219_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = string("op_2229_cast_fp16")]; + tensor var_2240_to_fp16 = const()[name = string("op_2240_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(236127616)))]; + tensor var_2241_to_fp16 = const()[name = string("op_2241_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(239404480)))]; + tensor linear_132_cast_fp16 = linear(bias = var_2241_to_fp16, weight = var_2240_to_fp16, x = var_2229_cast_fp16)[name = string("linear_132_cast_fp16")]; + tensor var_2244_to_fp16 = const()[name = string("op_2244_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(239407104)))]; + tensor linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2244_to_fp16, x = var_2229_cast_fp16)[name = string("linear_133_cast_fp16")]; + tensor var_2248_to_fp16 = const()[name = string("op_2248_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(242683968)))]; + tensor var_2249_to_fp16 = const()[name = string("op_2249_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(245960832)))]; + tensor linear_134_cast_fp16 = linear(bias = var_2249_to_fp16, weight = var_2248_to_fp16, x = var_2229_cast_fp16)[name = string("linear_134_cast_fp16")]; + tensor var_2253 = const()[name = string("op_2253"), val = tensor([1, 1500, 20, -1])]; + tensor var_2254_cast_fp16 = reshape(shape = var_2253, x = linear_132_cast_fp16)[name = string("op_2254_cast_fp16")]; + tensor var_2259 = const()[name = string("op_2259"), val = tensor([1, 1500, 20, -1])]; + tensor var_2260_cast_fp16 = reshape(shape = var_2259, x = linear_133_cast_fp16)[name = string("op_2260_cast_fp16")]; + tensor var_2265 = const()[name = string("op_2265"), val = tensor([1, 1500, 20, -1])]; + tensor var_2266_cast_fp16 = reshape(shape = var_2265, x = linear_134_cast_fp16)[name = string("op_2266_cast_fp16")]; + tensor transpose_114_perm_0 = const()[name = string("transpose_114_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_115_perm_0 = const()[name = string("transpose_115_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_116_perm_0 = const()[name = string("transpose_116_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_116 = transpose(perm = transpose_116_perm_0, x = var_2266_cast_fp16)[name = string("transpose_181")]; + tensor transpose_115 = transpose(perm = transpose_115_perm_0, x = var_2260_cast_fp16)[name = string("transpose_182")]; + tensor transpose_114 = transpose(perm = transpose_114_perm_0, x = var_2254_cast_fp16)[name = string("transpose_183")]; + tensor a_45_cast_fp16 = scaled_dot_product_attention(key = transpose_115, query = transpose_114, value = transpose_116)[name = string("a_45_cast_fp16")]; + tensor var_2270 = const()[name = string("op_2270"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = string("concat_22"), val = tensor([1, 1500, 1280])]; + tensor var_2271_cast_fp16 = transpose(perm = var_2270, x = a_45_cast_fp16)[name = string("transpose_180")]; + tensor x_275_cast_fp16 = reshape(shape = concat_22, x = var_2271_cast_fp16)[name = string("x_275_cast_fp16")]; + tensor var_2275_to_fp16 = const()[name = string("op_2275_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(245963456)))]; + tensor var_2276_to_fp16 = const()[name = string("op_2276_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(249240320)))]; + tensor linear_135_cast_fp16 = linear(bias = var_2276_to_fp16, weight = var_2275_to_fp16, x = x_275_cast_fp16)[name = string("linear_135_cast_fp16")]; + tensor x_277_cast_fp16 = add(x = x_271_cast_fp16, y = linear_135_cast_fp16)[name = string("x_277_cast_fp16")]; + tensor var_2283_axes_0 = const()[name = string("op_2283_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = string("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(249242944)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = string("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(249245568)))]; + tensor var_2283_cast_fp16 = layer_norm(axes = var_2283_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2219_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = string("op_2283_cast_fp16")]; + tensor var_2292_to_fp16 = const()[name = string("op_2292_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(249248192)))]; + tensor var_2293_to_fp16 = const()[name = string("op_2293_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(262355456)))]; + tensor linear_136_cast_fp16 = linear(bias = var_2293_to_fp16, weight = var_2292_to_fp16, x = var_2283_cast_fp16)[name = string("linear_136_cast_fp16")]; + string x_281_mode_0 = const()[name = string("x_281_mode_0"), val = string("EXACT")]; + tensor x_281_cast_fp16 = gelu(mode = x_281_mode_0, x = linear_136_cast_fp16)[name = string("x_281_cast_fp16")]; + tensor var_2298_to_fp16 = const()[name = string("op_2298_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(262365760)))]; + tensor var_2299_to_fp16 = const()[name = string("op_2299_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(275473024)))]; + tensor linear_137_cast_fp16 = linear(bias = var_2299_to_fp16, weight = var_2298_to_fp16, x = x_281_cast_fp16)[name = string("linear_137_cast_fp16")]; + tensor x_283_cast_fp16 = add(x = x_277_cast_fp16, y = linear_137_cast_fp16)[name = string("x_283_cast_fp16")]; + tensor var_2324_axes_0 = const()[name = string("op_2324_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = string("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(275475648)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = string("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(275478272)))]; + fp16 var_2314_to_fp16 = const()[name = string("op_2314_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2324_cast_fp16 = layer_norm(axes = var_2324_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2314_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = string("op_2324_cast_fp16")]; + tensor var_2335_to_fp16 = const()[name = string("op_2335_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(275480896)))]; + tensor var_2336_to_fp16 = const()[name = string("op_2336_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(278757760)))]; + tensor linear_138_cast_fp16 = linear(bias = var_2336_to_fp16, weight = var_2335_to_fp16, x = var_2324_cast_fp16)[name = string("linear_138_cast_fp16")]; + tensor var_2339_to_fp16 = const()[name = string("op_2339_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(278760384)))]; + tensor linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2339_to_fp16, x = var_2324_cast_fp16)[name = string("linear_139_cast_fp16")]; + tensor var_2343_to_fp16 = const()[name = string("op_2343_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(282037248)))]; + tensor var_2344_to_fp16 = const()[name = string("op_2344_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(285314112)))]; + tensor linear_140_cast_fp16 = linear(bias = var_2344_to_fp16, weight = var_2343_to_fp16, x = var_2324_cast_fp16)[name = string("linear_140_cast_fp16")]; + tensor var_2348 = const()[name = string("op_2348"), val = tensor([1, 1500, 20, -1])]; + tensor var_2349_cast_fp16 = reshape(shape = var_2348, x = linear_138_cast_fp16)[name = string("op_2349_cast_fp16")]; + tensor var_2354 = const()[name = string("op_2354"), val = tensor([1, 1500, 20, -1])]; + tensor var_2355_cast_fp16 = reshape(shape = var_2354, x = linear_139_cast_fp16)[name = string("op_2355_cast_fp16")]; + tensor var_2360 = const()[name = string("op_2360"), val = tensor([1, 1500, 20, -1])]; + tensor var_2361_cast_fp16 = reshape(shape = var_2360, x = linear_140_cast_fp16)[name = string("op_2361_cast_fp16")]; + tensor transpose_117_perm_0 = const()[name = string("transpose_117_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_118_perm_0 = const()[name = string("transpose_118_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_119_perm_0 = const()[name = string("transpose_119_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_119 = transpose(perm = transpose_119_perm_0, x = var_2361_cast_fp16)[name = string("transpose_177")]; + tensor transpose_118 = transpose(perm = transpose_118_perm_0, x = var_2355_cast_fp16)[name = string("transpose_178")]; + tensor transpose_117 = transpose(perm = transpose_117_perm_0, x = var_2349_cast_fp16)[name = string("transpose_179")]; + tensor a_47_cast_fp16 = scaled_dot_product_attention(key = transpose_118, query = transpose_117, value = transpose_119)[name = string("a_47_cast_fp16")]; + tensor var_2365 = const()[name = string("op_2365"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = string("concat_23"), val = tensor([1, 1500, 1280])]; + tensor var_2366_cast_fp16 = transpose(perm = var_2365, x = a_47_cast_fp16)[name = string("transpose_176")]; + tensor x_287_cast_fp16 = reshape(shape = concat_23, x = var_2366_cast_fp16)[name = string("x_287_cast_fp16")]; + tensor var_2370_to_fp16 = const()[name = string("op_2370_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(285316736)))]; + tensor var_2371_to_fp16 = const()[name = string("op_2371_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(288593600)))]; + tensor linear_141_cast_fp16 = linear(bias = var_2371_to_fp16, weight = var_2370_to_fp16, x = x_287_cast_fp16)[name = string("linear_141_cast_fp16")]; + tensor x_289_cast_fp16 = add(x = x_283_cast_fp16, y = linear_141_cast_fp16)[name = string("x_289_cast_fp16")]; + tensor var_2378_axes_0 = const()[name = string("op_2378_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = string("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(288596224)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = string("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(288598848)))]; + tensor var_2378_cast_fp16 = layer_norm(axes = var_2378_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2314_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = string("op_2378_cast_fp16")]; + tensor var_2387_to_fp16 = const()[name = string("op_2387_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(288601472)))]; + tensor var_2388_to_fp16 = const()[name = string("op_2388_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(301708736)))]; + tensor linear_142_cast_fp16 = linear(bias = var_2388_to_fp16, weight = var_2387_to_fp16, x = var_2378_cast_fp16)[name = string("linear_142_cast_fp16")]; + string x_293_mode_0 = const()[name = string("x_293_mode_0"), val = string("EXACT")]; + tensor x_293_cast_fp16 = gelu(mode = x_293_mode_0, x = linear_142_cast_fp16)[name = string("x_293_cast_fp16")]; + tensor var_2393_to_fp16 = const()[name = string("op_2393_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(301719040)))]; + tensor var_2394_to_fp16 = const()[name = string("op_2394_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(314826304)))]; + tensor linear_143_cast_fp16 = linear(bias = var_2394_to_fp16, weight = var_2393_to_fp16, x = x_293_cast_fp16)[name = string("linear_143_cast_fp16")]; + tensor x_295_cast_fp16 = add(x = x_289_cast_fp16, y = linear_143_cast_fp16)[name = string("x_295_cast_fp16")]; + tensor var_2419_axes_0 = const()[name = string("op_2419_axes_0"), val = tensor([-1])]; + tensor blocks_24_attn_ln_weight_to_fp16 = const()[name = string("blocks_24_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(314828928)))]; + tensor blocks_24_attn_ln_bias_to_fp16 = const()[name = string("blocks_24_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(314831552)))]; + fp16 var_2409_to_fp16 = const()[name = string("op_2409_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2419_cast_fp16 = layer_norm(axes = var_2419_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2409_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast_fp16)[name = string("op_2419_cast_fp16")]; + tensor var_2430_to_fp16 = const()[name = string("op_2430_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(314834176)))]; + tensor var_2431_to_fp16 = const()[name = string("op_2431_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(318111040)))]; + tensor linear_144_cast_fp16 = linear(bias = var_2431_to_fp16, weight = var_2430_to_fp16, x = var_2419_cast_fp16)[name = string("linear_144_cast_fp16")]; + tensor var_2434_to_fp16 = const()[name = string("op_2434_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(318113664)))]; + tensor linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2434_to_fp16, x = var_2419_cast_fp16)[name = string("linear_145_cast_fp16")]; + tensor var_2438_to_fp16 = const()[name = string("op_2438_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(321390528)))]; + tensor var_2439_to_fp16 = const()[name = string("op_2439_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(324667392)))]; + tensor linear_146_cast_fp16 = linear(bias = var_2439_to_fp16, weight = var_2438_to_fp16, x = var_2419_cast_fp16)[name = string("linear_146_cast_fp16")]; + tensor var_2443 = const()[name = string("op_2443"), val = tensor([1, 1500, 20, -1])]; + tensor var_2444_cast_fp16 = reshape(shape = var_2443, x = linear_144_cast_fp16)[name = string("op_2444_cast_fp16")]; + tensor var_2449 = const()[name = string("op_2449"), val = tensor([1, 1500, 20, -1])]; + tensor var_2450_cast_fp16 = reshape(shape = var_2449, x = linear_145_cast_fp16)[name = string("op_2450_cast_fp16")]; + tensor var_2455 = const()[name = string("op_2455"), val = tensor([1, 1500, 20, -1])]; + tensor var_2456_cast_fp16 = reshape(shape = var_2455, x = linear_146_cast_fp16)[name = string("op_2456_cast_fp16")]; + tensor transpose_120_perm_0 = const()[name = string("transpose_120_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_121_perm_0 = const()[name = string("transpose_121_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_122_perm_0 = const()[name = string("transpose_122_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_122 = transpose(perm = transpose_122_perm_0, x = var_2456_cast_fp16)[name = string("transpose_173")]; + tensor transpose_121 = transpose(perm = transpose_121_perm_0, x = var_2450_cast_fp16)[name = string("transpose_174")]; + tensor transpose_120 = transpose(perm = transpose_120_perm_0, x = var_2444_cast_fp16)[name = string("transpose_175")]; + tensor a_49_cast_fp16 = scaled_dot_product_attention(key = transpose_121, query = transpose_120, value = transpose_122)[name = string("a_49_cast_fp16")]; + tensor var_2460 = const()[name = string("op_2460"), val = tensor([0, 2, 1, 3])]; + tensor concat_24 = const()[name = string("concat_24"), val = tensor([1, 1500, 1280])]; + tensor var_2461_cast_fp16 = transpose(perm = var_2460, x = a_49_cast_fp16)[name = string("transpose_172")]; + tensor x_299_cast_fp16 = reshape(shape = concat_24, x = var_2461_cast_fp16)[name = string("x_299_cast_fp16")]; + tensor var_2465_to_fp16 = const()[name = string("op_2465_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(324670016)))]; + tensor var_2466_to_fp16 = const()[name = string("op_2466_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(327946880)))]; + tensor linear_147_cast_fp16 = linear(bias = var_2466_to_fp16, weight = var_2465_to_fp16, x = x_299_cast_fp16)[name = string("linear_147_cast_fp16")]; + tensor x_301_cast_fp16 = add(x = x_295_cast_fp16, y = linear_147_cast_fp16)[name = string("x_301_cast_fp16")]; + tensor var_2473_axes_0 = const()[name = string("op_2473_axes_0"), val = tensor([-1])]; + tensor blocks_24_mlp_ln_weight_to_fp16 = const()[name = string("blocks_24_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(327949504)))]; + tensor blocks_24_mlp_ln_bias_to_fp16 = const()[name = string("blocks_24_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(327952128)))]; + tensor var_2473_cast_fp16 = layer_norm(axes = var_2473_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2409_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast_fp16)[name = string("op_2473_cast_fp16")]; + tensor var_2482_to_fp16 = const()[name = string("op_2482_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(327954752)))]; + tensor var_2483_to_fp16 = const()[name = string("op_2483_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(341062016)))]; + tensor linear_148_cast_fp16 = linear(bias = var_2483_to_fp16, weight = var_2482_to_fp16, x = var_2473_cast_fp16)[name = string("linear_148_cast_fp16")]; + string x_305_mode_0 = const()[name = string("x_305_mode_0"), val = string("EXACT")]; + tensor x_305_cast_fp16 = gelu(mode = x_305_mode_0, x = linear_148_cast_fp16)[name = string("x_305_cast_fp16")]; + tensor var_2488_to_fp16 = const()[name = string("op_2488_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(341072320)))]; + tensor var_2489_to_fp16 = const()[name = string("op_2489_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(354179584)))]; + tensor linear_149_cast_fp16 = linear(bias = var_2489_to_fp16, weight = var_2488_to_fp16, x = x_305_cast_fp16)[name = string("linear_149_cast_fp16")]; + tensor x_307_cast_fp16 = add(x = x_301_cast_fp16, y = linear_149_cast_fp16)[name = string("x_307_cast_fp16")]; + tensor var_2514_axes_0 = const()[name = string("op_2514_axes_0"), val = tensor([-1])]; + tensor blocks_25_attn_ln_weight_to_fp16 = const()[name = string("blocks_25_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(354182208)))]; + tensor blocks_25_attn_ln_bias_to_fp16 = const()[name = string("blocks_25_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(354184832)))]; + fp16 var_2504_to_fp16 = const()[name = string("op_2504_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2514_cast_fp16 = layer_norm(axes = var_2514_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2504_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast_fp16)[name = string("op_2514_cast_fp16")]; + tensor var_2525_to_fp16 = const()[name = string("op_2525_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(354187456)))]; + tensor var_2526_to_fp16 = const()[name = string("op_2526_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(357464320)))]; + tensor linear_150_cast_fp16 = linear(bias = var_2526_to_fp16, weight = var_2525_to_fp16, x = var_2514_cast_fp16)[name = string("linear_150_cast_fp16")]; + tensor var_2529_to_fp16 = const()[name = string("op_2529_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(357466944)))]; + tensor linear_151_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2529_to_fp16, x = var_2514_cast_fp16)[name = string("linear_151_cast_fp16")]; + tensor var_2533_to_fp16 = const()[name = string("op_2533_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(360743808)))]; + tensor var_2534_to_fp16 = const()[name = string("op_2534_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(364020672)))]; + tensor linear_152_cast_fp16 = linear(bias = var_2534_to_fp16, weight = var_2533_to_fp16, x = var_2514_cast_fp16)[name = string("linear_152_cast_fp16")]; + tensor var_2538 = const()[name = string("op_2538"), val = tensor([1, 1500, 20, -1])]; + tensor var_2539_cast_fp16 = reshape(shape = var_2538, x = linear_150_cast_fp16)[name = string("op_2539_cast_fp16")]; + tensor var_2544 = const()[name = string("op_2544"), val = tensor([1, 1500, 20, -1])]; + tensor var_2545_cast_fp16 = reshape(shape = var_2544, x = linear_151_cast_fp16)[name = string("op_2545_cast_fp16")]; + tensor var_2550 = const()[name = string("op_2550"), val = tensor([1, 1500, 20, -1])]; + tensor var_2551_cast_fp16 = reshape(shape = var_2550, x = linear_152_cast_fp16)[name = string("op_2551_cast_fp16")]; + tensor transpose_123_perm_0 = const()[name = string("transpose_123_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_124_perm_0 = const()[name = string("transpose_124_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_125_perm_0 = const()[name = string("transpose_125_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_125 = transpose(perm = transpose_125_perm_0, x = var_2551_cast_fp16)[name = string("transpose_169")]; + tensor transpose_124 = transpose(perm = transpose_124_perm_0, x = var_2545_cast_fp16)[name = string("transpose_170")]; + tensor transpose_123 = transpose(perm = transpose_123_perm_0, x = var_2539_cast_fp16)[name = string("transpose_171")]; + tensor a_51_cast_fp16 = scaled_dot_product_attention(key = transpose_124, query = transpose_123, value = transpose_125)[name = string("a_51_cast_fp16")]; + tensor var_2555 = const()[name = string("op_2555"), val = tensor([0, 2, 1, 3])]; + tensor concat_25 = const()[name = string("concat_25"), val = tensor([1, 1500, 1280])]; + tensor var_2556_cast_fp16 = transpose(perm = var_2555, x = a_51_cast_fp16)[name = string("transpose_168")]; + tensor x_311_cast_fp16 = reshape(shape = concat_25, x = var_2556_cast_fp16)[name = string("x_311_cast_fp16")]; + tensor var_2560_to_fp16 = const()[name = string("op_2560_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(364023296)))]; + tensor var_2561_to_fp16 = const()[name = string("op_2561_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(367300160)))]; + tensor linear_153_cast_fp16 = linear(bias = var_2561_to_fp16, weight = var_2560_to_fp16, x = x_311_cast_fp16)[name = string("linear_153_cast_fp16")]; + tensor x_313_cast_fp16 = add(x = x_307_cast_fp16, y = linear_153_cast_fp16)[name = string("x_313_cast_fp16")]; + tensor var_2568_axes_0 = const()[name = string("op_2568_axes_0"), val = tensor([-1])]; + tensor blocks_25_mlp_ln_weight_to_fp16 = const()[name = string("blocks_25_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(367302784)))]; + tensor blocks_25_mlp_ln_bias_to_fp16 = const()[name = string("blocks_25_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(367305408)))]; + tensor var_2568_cast_fp16 = layer_norm(axes = var_2568_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2504_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast_fp16)[name = string("op_2568_cast_fp16")]; + tensor var_2577_to_fp16 = const()[name = string("op_2577_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(367308032)))]; + tensor var_2578_to_fp16 = const()[name = string("op_2578_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(380415296)))]; + tensor linear_154_cast_fp16 = linear(bias = var_2578_to_fp16, weight = var_2577_to_fp16, x = var_2568_cast_fp16)[name = string("linear_154_cast_fp16")]; + string x_317_mode_0 = const()[name = string("x_317_mode_0"), val = string("EXACT")]; + tensor x_317_cast_fp16 = gelu(mode = x_317_mode_0, x = linear_154_cast_fp16)[name = string("x_317_cast_fp16")]; + tensor var_2583_to_fp16 = const()[name = string("op_2583_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(380425600)))]; + tensor var_2584_to_fp16 = const()[name = string("op_2584_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(393532864)))]; + tensor linear_155_cast_fp16 = linear(bias = var_2584_to_fp16, weight = var_2583_to_fp16, x = x_317_cast_fp16)[name = string("linear_155_cast_fp16")]; + tensor x_319_cast_fp16 = add(x = x_313_cast_fp16, y = linear_155_cast_fp16)[name = string("x_319_cast_fp16")]; + tensor var_2609_axes_0 = const()[name = string("op_2609_axes_0"), val = tensor([-1])]; + tensor blocks_26_attn_ln_weight_to_fp16 = const()[name = string("blocks_26_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(393535488)))]; + tensor blocks_26_attn_ln_bias_to_fp16 = const()[name = string("blocks_26_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(393538112)))]; + fp16 var_2599_to_fp16 = const()[name = string("op_2599_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2609_cast_fp16 = layer_norm(axes = var_2609_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2599_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast_fp16)[name = string("op_2609_cast_fp16")]; + tensor var_2620_to_fp16 = const()[name = string("op_2620_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(393540736)))]; + tensor var_2621_to_fp16 = const()[name = string("op_2621_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(396817600)))]; + tensor linear_156_cast_fp16 = linear(bias = var_2621_to_fp16, weight = var_2620_to_fp16, x = var_2609_cast_fp16)[name = string("linear_156_cast_fp16")]; + tensor var_2624_to_fp16 = const()[name = string("op_2624_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(396820224)))]; + tensor linear_157_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2624_to_fp16, x = var_2609_cast_fp16)[name = string("linear_157_cast_fp16")]; + tensor var_2628_to_fp16 = const()[name = string("op_2628_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(400097088)))]; + tensor var_2629_to_fp16 = const()[name = string("op_2629_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(403373952)))]; + tensor linear_158_cast_fp16 = linear(bias = var_2629_to_fp16, weight = var_2628_to_fp16, x = var_2609_cast_fp16)[name = string("linear_158_cast_fp16")]; + tensor var_2633 = const()[name = string("op_2633"), val = tensor([1, 1500, 20, -1])]; + tensor var_2634_cast_fp16 = reshape(shape = var_2633, x = linear_156_cast_fp16)[name = string("op_2634_cast_fp16")]; + tensor var_2639 = const()[name = string("op_2639"), val = tensor([1, 1500, 20, -1])]; + tensor var_2640_cast_fp16 = reshape(shape = var_2639, x = linear_157_cast_fp16)[name = string("op_2640_cast_fp16")]; + tensor var_2645 = const()[name = string("op_2645"), val = tensor([1, 1500, 20, -1])]; + tensor var_2646_cast_fp16 = reshape(shape = var_2645, x = linear_158_cast_fp16)[name = string("op_2646_cast_fp16")]; + tensor transpose_126_perm_0 = const()[name = string("transpose_126_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_127_perm_0 = const()[name = string("transpose_127_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_128 = transpose(perm = transpose_128_perm_0, x = var_2646_cast_fp16)[name = string("transpose_165")]; + tensor transpose_127 = transpose(perm = transpose_127_perm_0, x = var_2640_cast_fp16)[name = string("transpose_166")]; + tensor transpose_126 = transpose(perm = transpose_126_perm_0, x = var_2634_cast_fp16)[name = string("transpose_167")]; + tensor a_53_cast_fp16 = scaled_dot_product_attention(key = transpose_127, query = transpose_126, value = transpose_128)[name = string("a_53_cast_fp16")]; + tensor var_2650 = const()[name = string("op_2650"), val = tensor([0, 2, 1, 3])]; + tensor concat_26 = const()[name = string("concat_26"), val = tensor([1, 1500, 1280])]; + tensor var_2651_cast_fp16 = transpose(perm = var_2650, x = a_53_cast_fp16)[name = string("transpose_164")]; + tensor x_323_cast_fp16 = reshape(shape = concat_26, x = var_2651_cast_fp16)[name = string("x_323_cast_fp16")]; + tensor var_2655_to_fp16 = const()[name = string("op_2655_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(403376576)))]; + tensor var_2656_to_fp16 = const()[name = string("op_2656_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(406653440)))]; + tensor linear_159_cast_fp16 = linear(bias = var_2656_to_fp16, weight = var_2655_to_fp16, x = x_323_cast_fp16)[name = string("linear_159_cast_fp16")]; + tensor x_325_cast_fp16 = add(x = x_319_cast_fp16, y = linear_159_cast_fp16)[name = string("x_325_cast_fp16")]; + tensor var_2663_axes_0 = const()[name = string("op_2663_axes_0"), val = tensor([-1])]; + tensor blocks_26_mlp_ln_weight_to_fp16 = const()[name = string("blocks_26_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(406656064)))]; + tensor blocks_26_mlp_ln_bias_to_fp16 = const()[name = string("blocks_26_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(406658688)))]; + tensor var_2663_cast_fp16 = layer_norm(axes = var_2663_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2599_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast_fp16)[name = string("op_2663_cast_fp16")]; + tensor var_2672_to_fp16 = const()[name = string("op_2672_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(406661312)))]; + tensor var_2673_to_fp16 = const()[name = string("op_2673_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(419768576)))]; + tensor linear_160_cast_fp16 = linear(bias = var_2673_to_fp16, weight = var_2672_to_fp16, x = var_2663_cast_fp16)[name = string("linear_160_cast_fp16")]; + string x_329_mode_0 = const()[name = string("x_329_mode_0"), val = string("EXACT")]; + tensor x_329_cast_fp16 = gelu(mode = x_329_mode_0, x = linear_160_cast_fp16)[name = string("x_329_cast_fp16")]; + tensor var_2678_to_fp16 = const()[name = string("op_2678_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(419778880)))]; + tensor var_2679_to_fp16 = const()[name = string("op_2679_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(432886144)))]; + tensor linear_161_cast_fp16 = linear(bias = var_2679_to_fp16, weight = var_2678_to_fp16, x = x_329_cast_fp16)[name = string("linear_161_cast_fp16")]; + tensor x_331_cast_fp16 = add(x = x_325_cast_fp16, y = linear_161_cast_fp16)[name = string("x_331_cast_fp16")]; + tensor var_2704_axes_0 = const()[name = string("op_2704_axes_0"), val = tensor([-1])]; + tensor blocks_27_attn_ln_weight_to_fp16 = const()[name = string("blocks_27_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(432888768)))]; + tensor blocks_27_attn_ln_bias_to_fp16 = const()[name = string("blocks_27_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(432891392)))]; + fp16 var_2694_to_fp16 = const()[name = string("op_2694_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2704_cast_fp16 = layer_norm(axes = var_2704_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_2694_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast_fp16)[name = string("op_2704_cast_fp16")]; + tensor var_2715_to_fp16 = const()[name = string("op_2715_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(432894016)))]; + tensor var_2716_to_fp16 = const()[name = string("op_2716_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(436170880)))]; + tensor linear_162_cast_fp16 = linear(bias = var_2716_to_fp16, weight = var_2715_to_fp16, x = var_2704_cast_fp16)[name = string("linear_162_cast_fp16")]; + tensor var_2719_to_fp16 = const()[name = string("op_2719_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(436173504)))]; + tensor linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2719_to_fp16, x = var_2704_cast_fp16)[name = string("linear_163_cast_fp16")]; + tensor var_2723_to_fp16 = const()[name = string("op_2723_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(439450368)))]; + tensor var_2724_to_fp16 = const()[name = string("op_2724_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(442727232)))]; + tensor linear_164_cast_fp16 = linear(bias = var_2724_to_fp16, weight = var_2723_to_fp16, x = var_2704_cast_fp16)[name = string("linear_164_cast_fp16")]; + tensor var_2728 = const()[name = string("op_2728"), val = tensor([1, 1500, 20, -1])]; + tensor var_2729_cast_fp16 = reshape(shape = var_2728, x = linear_162_cast_fp16)[name = string("op_2729_cast_fp16")]; + tensor var_2734 = const()[name = string("op_2734"), val = tensor([1, 1500, 20, -1])]; + tensor var_2735_cast_fp16 = reshape(shape = var_2734, x = linear_163_cast_fp16)[name = string("op_2735_cast_fp16")]; + tensor var_2740 = const()[name = string("op_2740"), val = tensor([1, 1500, 20, -1])]; + tensor var_2741_cast_fp16 = reshape(shape = var_2740, x = linear_164_cast_fp16)[name = string("op_2741_cast_fp16")]; + tensor transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_131 = transpose(perm = transpose_131_perm_0, x = var_2741_cast_fp16)[name = string("transpose_161")]; + tensor transpose_130 = transpose(perm = transpose_130_perm_0, x = var_2735_cast_fp16)[name = string("transpose_162")]; + tensor transpose_129 = transpose(perm = transpose_129_perm_0, x = var_2729_cast_fp16)[name = string("transpose_163")]; + tensor a_55_cast_fp16 = scaled_dot_product_attention(key = transpose_130, query = transpose_129, value = transpose_131)[name = string("a_55_cast_fp16")]; + tensor var_2745 = const()[name = string("op_2745"), val = tensor([0, 2, 1, 3])]; + tensor concat_27 = const()[name = string("concat_27"), val = tensor([1, 1500, 1280])]; + tensor var_2746_cast_fp16 = transpose(perm = var_2745, x = a_55_cast_fp16)[name = string("transpose_160")]; + tensor x_335_cast_fp16 = reshape(shape = concat_27, x = var_2746_cast_fp16)[name = string("x_335_cast_fp16")]; + tensor var_2750_to_fp16 = const()[name = string("op_2750_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(442729856)))]; + tensor var_2751_to_fp16 = const()[name = string("op_2751_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(446006720)))]; + tensor linear_165_cast_fp16 = linear(bias = var_2751_to_fp16, weight = var_2750_to_fp16, x = x_335_cast_fp16)[name = string("linear_165_cast_fp16")]; + tensor x_337_cast_fp16 = add(x = x_331_cast_fp16, y = linear_165_cast_fp16)[name = string("x_337_cast_fp16")]; + tensor var_2758_axes_0 = const()[name = string("op_2758_axes_0"), val = tensor([-1])]; + tensor blocks_27_mlp_ln_weight_to_fp16 = const()[name = string("blocks_27_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(446009344)))]; + tensor blocks_27_mlp_ln_bias_to_fp16 = const()[name = string("blocks_27_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(446011968)))]; + tensor var_2758_cast_fp16 = layer_norm(axes = var_2758_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_2694_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast_fp16)[name = string("op_2758_cast_fp16")]; + tensor var_2767_to_fp16 = const()[name = string("op_2767_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(446014592)))]; + tensor var_2768_to_fp16 = const()[name = string("op_2768_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(459121856)))]; + tensor linear_166_cast_fp16 = linear(bias = var_2768_to_fp16, weight = var_2767_to_fp16, x = var_2758_cast_fp16)[name = string("linear_166_cast_fp16")]; + string x_341_mode_0 = const()[name = string("x_341_mode_0"), val = string("EXACT")]; + tensor x_341_cast_fp16 = gelu(mode = x_341_mode_0, x = linear_166_cast_fp16)[name = string("x_341_cast_fp16")]; + tensor var_2773_to_fp16 = const()[name = string("op_2773_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(459132160)))]; + tensor var_2774_to_fp16 = const()[name = string("op_2774_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(472239424)))]; + tensor linear_167_cast_fp16 = linear(bias = var_2774_to_fp16, weight = var_2773_to_fp16, x = x_341_cast_fp16)[name = string("linear_167_cast_fp16")]; + tensor x_343_cast_fp16 = add(x = x_337_cast_fp16, y = linear_167_cast_fp16)[name = string("x_343_cast_fp16")]; + tensor var_2799_axes_0 = const()[name = string("op_2799_axes_0"), val = tensor([-1])]; + tensor blocks_28_attn_ln_weight_to_fp16 = const()[name = string("blocks_28_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(472242048)))]; + tensor blocks_28_attn_ln_bias_to_fp16 = const()[name = string("blocks_28_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(472244672)))]; + fp16 var_2789_to_fp16 = const()[name = string("op_2789_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2799_cast_fp16 = layer_norm(axes = var_2799_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_2789_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast_fp16)[name = string("op_2799_cast_fp16")]; + tensor var_2810_to_fp16 = const()[name = string("op_2810_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(472247296)))]; + tensor var_2811_to_fp16 = const()[name = string("op_2811_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(475524160)))]; + tensor linear_168_cast_fp16 = linear(bias = var_2811_to_fp16, weight = var_2810_to_fp16, x = var_2799_cast_fp16)[name = string("linear_168_cast_fp16")]; + tensor var_2814_to_fp16 = const()[name = string("op_2814_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(475526784)))]; + tensor linear_169_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2814_to_fp16, x = var_2799_cast_fp16)[name = string("linear_169_cast_fp16")]; + tensor var_2818_to_fp16 = const()[name = string("op_2818_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(478803648)))]; + tensor var_2819_to_fp16 = const()[name = string("op_2819_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(482080512)))]; + tensor linear_170_cast_fp16 = linear(bias = var_2819_to_fp16, weight = var_2818_to_fp16, x = var_2799_cast_fp16)[name = string("linear_170_cast_fp16")]; + tensor var_2823 = const()[name = string("op_2823"), val = tensor([1, 1500, 20, -1])]; + tensor var_2824_cast_fp16 = reshape(shape = var_2823, x = linear_168_cast_fp16)[name = string("op_2824_cast_fp16")]; + tensor var_2829 = const()[name = string("op_2829"), val = tensor([1, 1500, 20, -1])]; + tensor var_2830_cast_fp16 = reshape(shape = var_2829, x = linear_169_cast_fp16)[name = string("op_2830_cast_fp16")]; + tensor var_2835 = const()[name = string("op_2835"), val = tensor([1, 1500, 20, -1])]; + tensor var_2836_cast_fp16 = reshape(shape = var_2835, x = linear_170_cast_fp16)[name = string("op_2836_cast_fp16")]; + tensor transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_134 = transpose(perm = transpose_134_perm_0, x = var_2836_cast_fp16)[name = string("transpose_157")]; + tensor transpose_133 = transpose(perm = transpose_133_perm_0, x = var_2830_cast_fp16)[name = string("transpose_158")]; + tensor transpose_132 = transpose(perm = transpose_132_perm_0, x = var_2824_cast_fp16)[name = string("transpose_159")]; + tensor a_57_cast_fp16 = scaled_dot_product_attention(key = transpose_133, query = transpose_132, value = transpose_134)[name = string("a_57_cast_fp16")]; + tensor var_2840 = const()[name = string("op_2840"), val = tensor([0, 2, 1, 3])]; + tensor concat_28 = const()[name = string("concat_28"), val = tensor([1, 1500, 1280])]; + tensor var_2841_cast_fp16 = transpose(perm = var_2840, x = a_57_cast_fp16)[name = string("transpose_156")]; + tensor x_347_cast_fp16 = reshape(shape = concat_28, x = var_2841_cast_fp16)[name = string("x_347_cast_fp16")]; + tensor var_2845_to_fp16 = const()[name = string("op_2845_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(482083136)))]; + tensor var_2846_to_fp16 = const()[name = string("op_2846_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(485360000)))]; + tensor linear_171_cast_fp16 = linear(bias = var_2846_to_fp16, weight = var_2845_to_fp16, x = x_347_cast_fp16)[name = string("linear_171_cast_fp16")]; + tensor x_349_cast_fp16 = add(x = x_343_cast_fp16, y = linear_171_cast_fp16)[name = string("x_349_cast_fp16")]; + tensor var_2853_axes_0 = const()[name = string("op_2853_axes_0"), val = tensor([-1])]; + tensor blocks_28_mlp_ln_weight_to_fp16 = const()[name = string("blocks_28_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(485362624)))]; + tensor blocks_28_mlp_ln_bias_to_fp16 = const()[name = string("blocks_28_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(485365248)))]; + tensor var_2853_cast_fp16 = layer_norm(axes = var_2853_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_2789_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast_fp16)[name = string("op_2853_cast_fp16")]; + tensor var_2862_to_fp16 = const()[name = string("op_2862_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(485367872)))]; + tensor var_2863_to_fp16 = const()[name = string("op_2863_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(498475136)))]; + tensor linear_172_cast_fp16 = linear(bias = var_2863_to_fp16, weight = var_2862_to_fp16, x = var_2853_cast_fp16)[name = string("linear_172_cast_fp16")]; + string x_353_mode_0 = const()[name = string("x_353_mode_0"), val = string("EXACT")]; + tensor x_353_cast_fp16 = gelu(mode = x_353_mode_0, x = linear_172_cast_fp16)[name = string("x_353_cast_fp16")]; + tensor var_2868_to_fp16 = const()[name = string("op_2868_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(498485440)))]; + tensor var_2869_to_fp16 = const()[name = string("op_2869_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(511592704)))]; + tensor linear_173_cast_fp16 = linear(bias = var_2869_to_fp16, weight = var_2868_to_fp16, x = x_353_cast_fp16)[name = string("linear_173_cast_fp16")]; + tensor x_355_cast_fp16 = add(x = x_349_cast_fp16, y = linear_173_cast_fp16)[name = string("x_355_cast_fp16")]; + tensor var_2894_axes_0 = const()[name = string("op_2894_axes_0"), val = tensor([-1])]; + tensor blocks_29_attn_ln_weight_to_fp16 = const()[name = string("blocks_29_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(511595328)))]; + tensor blocks_29_attn_ln_bias_to_fp16 = const()[name = string("blocks_29_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(511597952)))]; + fp16 var_2884_to_fp16 = const()[name = string("op_2884_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2894_cast_fp16 = layer_norm(axes = var_2894_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_2884_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast_fp16)[name = string("op_2894_cast_fp16")]; + tensor var_2905_to_fp16 = const()[name = string("op_2905_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(511600576)))]; + tensor var_2906_to_fp16 = const()[name = string("op_2906_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(514877440)))]; + tensor linear_174_cast_fp16 = linear(bias = var_2906_to_fp16, weight = var_2905_to_fp16, x = var_2894_cast_fp16)[name = string("linear_174_cast_fp16")]; + tensor var_2909_to_fp16 = const()[name = string("op_2909_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(514880064)))]; + tensor linear_175_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2909_to_fp16, x = var_2894_cast_fp16)[name = string("linear_175_cast_fp16")]; + tensor var_2913_to_fp16 = const()[name = string("op_2913_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(518156928)))]; + tensor var_2914_to_fp16 = const()[name = string("op_2914_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(521433792)))]; + tensor linear_176_cast_fp16 = linear(bias = var_2914_to_fp16, weight = var_2913_to_fp16, x = var_2894_cast_fp16)[name = string("linear_176_cast_fp16")]; + tensor var_2918 = const()[name = string("op_2918"), val = tensor([1, 1500, 20, -1])]; + tensor var_2919_cast_fp16 = reshape(shape = var_2918, x = linear_174_cast_fp16)[name = string("op_2919_cast_fp16")]; + tensor var_2924 = const()[name = string("op_2924"), val = tensor([1, 1500, 20, -1])]; + tensor var_2925_cast_fp16 = reshape(shape = var_2924, x = linear_175_cast_fp16)[name = string("op_2925_cast_fp16")]; + tensor var_2930 = const()[name = string("op_2930"), val = tensor([1, 1500, 20, -1])]; + tensor var_2931_cast_fp16 = reshape(shape = var_2930, x = linear_176_cast_fp16)[name = string("op_2931_cast_fp16")]; + tensor transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_137 = transpose(perm = transpose_137_perm_0, x = var_2931_cast_fp16)[name = string("transpose_153")]; + tensor transpose_136 = transpose(perm = transpose_136_perm_0, x = var_2925_cast_fp16)[name = string("transpose_154")]; + tensor transpose_135 = transpose(perm = transpose_135_perm_0, x = var_2919_cast_fp16)[name = string("transpose_155")]; + tensor a_59_cast_fp16 = scaled_dot_product_attention(key = transpose_136, query = transpose_135, value = transpose_137)[name = string("a_59_cast_fp16")]; + tensor var_2935 = const()[name = string("op_2935"), val = tensor([0, 2, 1, 3])]; + tensor concat_29 = const()[name = string("concat_29"), val = tensor([1, 1500, 1280])]; + tensor var_2936_cast_fp16 = transpose(perm = var_2935, x = a_59_cast_fp16)[name = string("transpose_152")]; + tensor x_359_cast_fp16 = reshape(shape = concat_29, x = var_2936_cast_fp16)[name = string("x_359_cast_fp16")]; + tensor var_2940_to_fp16 = const()[name = string("op_2940_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(521436416)))]; + tensor var_2941_to_fp16 = const()[name = string("op_2941_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(524713280)))]; + tensor linear_177_cast_fp16 = linear(bias = var_2941_to_fp16, weight = var_2940_to_fp16, x = x_359_cast_fp16)[name = string("linear_177_cast_fp16")]; + tensor x_361_cast_fp16 = add(x = x_355_cast_fp16, y = linear_177_cast_fp16)[name = string("x_361_cast_fp16")]; + tensor var_2948_axes_0 = const()[name = string("op_2948_axes_0"), val = tensor([-1])]; + tensor blocks_29_mlp_ln_weight_to_fp16 = const()[name = string("blocks_29_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(524715904)))]; + tensor blocks_29_mlp_ln_bias_to_fp16 = const()[name = string("blocks_29_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(524718528)))]; + tensor var_2948_cast_fp16 = layer_norm(axes = var_2948_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_2884_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast_fp16)[name = string("op_2948_cast_fp16")]; + tensor var_2957_to_fp16 = const()[name = string("op_2957_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(524721152)))]; + tensor var_2958_to_fp16 = const()[name = string("op_2958_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(537828416)))]; + tensor linear_178_cast_fp16 = linear(bias = var_2958_to_fp16, weight = var_2957_to_fp16, x = var_2948_cast_fp16)[name = string("linear_178_cast_fp16")]; + string x_365_mode_0 = const()[name = string("x_365_mode_0"), val = string("EXACT")]; + tensor x_365_cast_fp16 = gelu(mode = x_365_mode_0, x = linear_178_cast_fp16)[name = string("x_365_cast_fp16")]; + tensor var_2963_to_fp16 = const()[name = string("op_2963_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(537838720)))]; + tensor var_2964_to_fp16 = const()[name = string("op_2964_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(550945984)))]; + tensor linear_179_cast_fp16 = linear(bias = var_2964_to_fp16, weight = var_2963_to_fp16, x = x_365_cast_fp16)[name = string("linear_179_cast_fp16")]; + tensor x_367_cast_fp16 = add(x = x_361_cast_fp16, y = linear_179_cast_fp16)[name = string("x_367_cast_fp16")]; + tensor var_2989_axes_0 = const()[name = string("op_2989_axes_0"), val = tensor([-1])]; + tensor blocks_30_attn_ln_weight_to_fp16 = const()[name = string("blocks_30_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(550948608)))]; + tensor blocks_30_attn_ln_bias_to_fp16 = const()[name = string("blocks_30_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(550951232)))]; + fp16 var_2979_to_fp16 = const()[name = string("op_2979_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_2989_cast_fp16 = layer_norm(axes = var_2989_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_2979_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast_fp16)[name = string("op_2989_cast_fp16")]; + tensor var_3000_to_fp16 = const()[name = string("op_3000_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(550953856)))]; + tensor var_3001_to_fp16 = const()[name = string("op_3001_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(554230720)))]; + tensor linear_180_cast_fp16 = linear(bias = var_3001_to_fp16, weight = var_3000_to_fp16, x = var_2989_cast_fp16)[name = string("linear_180_cast_fp16")]; + tensor var_3004_to_fp16 = const()[name = string("op_3004_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(554233344)))]; + tensor linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3004_to_fp16, x = var_2989_cast_fp16)[name = string("linear_181_cast_fp16")]; + tensor var_3008_to_fp16 = const()[name = string("op_3008_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(557510208)))]; + tensor var_3009_to_fp16 = const()[name = string("op_3009_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(560787072)))]; + tensor linear_182_cast_fp16 = linear(bias = var_3009_to_fp16, weight = var_3008_to_fp16, x = var_2989_cast_fp16)[name = string("linear_182_cast_fp16")]; + tensor var_3013 = const()[name = string("op_3013"), val = tensor([1, 1500, 20, -1])]; + tensor var_3014_cast_fp16 = reshape(shape = var_3013, x = linear_180_cast_fp16)[name = string("op_3014_cast_fp16")]; + tensor var_3019 = const()[name = string("op_3019"), val = tensor([1, 1500, 20, -1])]; + tensor var_3020_cast_fp16 = reshape(shape = var_3019, x = linear_181_cast_fp16)[name = string("op_3020_cast_fp16")]; + tensor var_3025 = const()[name = string("op_3025"), val = tensor([1, 1500, 20, -1])]; + tensor var_3026_cast_fp16 = reshape(shape = var_3025, x = linear_182_cast_fp16)[name = string("op_3026_cast_fp16")]; + tensor transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_140 = transpose(perm = transpose_140_perm_0, x = var_3026_cast_fp16)[name = string("transpose_149")]; + tensor transpose_139 = transpose(perm = transpose_139_perm_0, x = var_3020_cast_fp16)[name = string("transpose_150")]; + tensor transpose_138 = transpose(perm = transpose_138_perm_0, x = var_3014_cast_fp16)[name = string("transpose_151")]; + tensor a_61_cast_fp16 = scaled_dot_product_attention(key = transpose_139, query = transpose_138, value = transpose_140)[name = string("a_61_cast_fp16")]; + tensor var_3030 = const()[name = string("op_3030"), val = tensor([0, 2, 1, 3])]; + tensor concat_30 = const()[name = string("concat_30"), val = tensor([1, 1500, 1280])]; + tensor var_3031_cast_fp16 = transpose(perm = var_3030, x = a_61_cast_fp16)[name = string("transpose_148")]; + tensor x_371_cast_fp16 = reshape(shape = concat_30, x = var_3031_cast_fp16)[name = string("x_371_cast_fp16")]; + tensor var_3035_to_fp16 = const()[name = string("op_3035_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(560789696)))]; + tensor var_3036_to_fp16 = const()[name = string("op_3036_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(564066560)))]; + tensor linear_183_cast_fp16 = linear(bias = var_3036_to_fp16, weight = var_3035_to_fp16, x = x_371_cast_fp16)[name = string("linear_183_cast_fp16")]; + tensor x_373_cast_fp16 = add(x = x_367_cast_fp16, y = linear_183_cast_fp16)[name = string("x_373_cast_fp16")]; + tensor var_3043_axes_0 = const()[name = string("op_3043_axes_0"), val = tensor([-1])]; + tensor blocks_30_mlp_ln_weight_to_fp16 = const()[name = string("blocks_30_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(564069184)))]; + tensor blocks_30_mlp_ln_bias_to_fp16 = const()[name = string("blocks_30_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(564071808)))]; + tensor var_3043_cast_fp16 = layer_norm(axes = var_3043_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_2979_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast_fp16)[name = string("op_3043_cast_fp16")]; + tensor var_3052_to_fp16 = const()[name = string("op_3052_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(564074432)))]; + tensor var_3053_to_fp16 = const()[name = string("op_3053_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(577181696)))]; + tensor linear_184_cast_fp16 = linear(bias = var_3053_to_fp16, weight = var_3052_to_fp16, x = var_3043_cast_fp16)[name = string("linear_184_cast_fp16")]; + string x_377_mode_0 = const()[name = string("x_377_mode_0"), val = string("EXACT")]; + tensor x_377_cast_fp16 = gelu(mode = x_377_mode_0, x = linear_184_cast_fp16)[name = string("x_377_cast_fp16")]; + tensor var_3058_to_fp16 = const()[name = string("op_3058_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(577192000)))]; + tensor var_3059_to_fp16 = const()[name = string("op_3059_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(590299264)))]; + tensor linear_185_cast_fp16 = linear(bias = var_3059_to_fp16, weight = var_3058_to_fp16, x = x_377_cast_fp16)[name = string("linear_185_cast_fp16")]; + tensor x_379_cast_fp16 = add(x = x_373_cast_fp16, y = linear_185_cast_fp16)[name = string("x_379_cast_fp16")]; + tensor var_3084_axes_0 = const()[name = string("op_3084_axes_0"), val = tensor([-1])]; + tensor blocks_31_attn_ln_weight_to_fp16 = const()[name = string("blocks_31_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(590301888)))]; + tensor blocks_31_attn_ln_bias_to_fp16 = const()[name = string("blocks_31_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(590304512)))]; + fp16 var_3074_to_fp16 = const()[name = string("op_3074_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_3084_cast_fp16 = layer_norm(axes = var_3084_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3074_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast_fp16)[name = string("op_3084_cast_fp16")]; + tensor var_3095_to_fp16 = const()[name = string("op_3095_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(590307136)))]; + tensor var_3096_to_fp16 = const()[name = string("op_3096_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(593584000)))]; + tensor linear_186_cast_fp16 = linear(bias = var_3096_to_fp16, weight = var_3095_to_fp16, x = var_3084_cast_fp16)[name = string("linear_186_cast_fp16")]; + tensor var_3099_to_fp16 = const()[name = string("op_3099_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(593586624)))]; + tensor linear_187_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3099_to_fp16, x = var_3084_cast_fp16)[name = string("linear_187_cast_fp16")]; + tensor var_3103_to_fp16 = const()[name = string("op_3103_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(596863488)))]; + tensor var_3104_to_fp16 = const()[name = string("op_3104_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(600140352)))]; + tensor linear_188_cast_fp16 = linear(bias = var_3104_to_fp16, weight = var_3103_to_fp16, x = var_3084_cast_fp16)[name = string("linear_188_cast_fp16")]; + tensor var_3108 = const()[name = string("op_3108"), val = tensor([1, 1500, 20, -1])]; + tensor var_3109_cast_fp16 = reshape(shape = var_3108, x = linear_186_cast_fp16)[name = string("op_3109_cast_fp16")]; + tensor var_3114 = const()[name = string("op_3114"), val = tensor([1, 1500, 20, -1])]; + tensor var_3115_cast_fp16 = reshape(shape = var_3114, x = linear_187_cast_fp16)[name = string("op_3115_cast_fp16")]; + tensor var_3120 = const()[name = string("op_3120"), val = tensor([1, 1500, 20, -1])]; + tensor var_3121_cast_fp16 = reshape(shape = var_3120, x = linear_188_cast_fp16)[name = string("op_3121_cast_fp16")]; + tensor transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_143 = transpose(perm = transpose_143_perm_0, x = var_3121_cast_fp16)[name = string("transpose_145")]; + tensor transpose_142 = transpose(perm = transpose_142_perm_0, x = var_3115_cast_fp16)[name = string("transpose_146")]; + tensor transpose_141 = transpose(perm = transpose_141_perm_0, x = var_3109_cast_fp16)[name = string("transpose_147")]; + tensor a_cast_fp16 = scaled_dot_product_attention(key = transpose_142, query = transpose_141, value = transpose_143)[name = string("a_cast_fp16")]; + tensor var_3125 = const()[name = string("op_3125"), val = tensor([0, 2, 1, 3])]; + tensor concat_31 = const()[name = string("concat_31"), val = tensor([1, 1500, 1280])]; + tensor var_3126_cast_fp16 = transpose(perm = var_3125, x = a_cast_fp16)[name = string("transpose_144")]; + tensor x_383_cast_fp16 = reshape(shape = concat_31, x = var_3126_cast_fp16)[name = string("x_383_cast_fp16")]; + tensor var_3130_to_fp16 = const()[name = string("op_3130_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(600142976)))]; + tensor var_3131_to_fp16 = const()[name = string("op_3131_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(603419840)))]; + tensor linear_189_cast_fp16 = linear(bias = var_3131_to_fp16, weight = var_3130_to_fp16, x = x_383_cast_fp16)[name = string("linear_189_cast_fp16")]; + tensor x_385_cast_fp16 = add(x = x_379_cast_fp16, y = linear_189_cast_fp16)[name = string("x_385_cast_fp16")]; + tensor var_3138_axes_0 = const()[name = string("op_3138_axes_0"), val = tensor([-1])]; + tensor blocks_31_mlp_ln_weight_to_fp16 = const()[name = string("blocks_31_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(603422464)))]; + tensor blocks_31_mlp_ln_bias_to_fp16 = const()[name = string("blocks_31_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(603425088)))]; + tensor var_3138_cast_fp16 = layer_norm(axes = var_3138_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3074_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast_fp16)[name = string("op_3138_cast_fp16")]; + tensor var_3147_to_fp16 = const()[name = string("op_3147_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(603427712)))]; + tensor var_3148_to_fp16 = const()[name = string("op_3148_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(616534976)))]; + tensor linear_190_cast_fp16 = linear(bias = var_3148_to_fp16, weight = var_3147_to_fp16, x = var_3138_cast_fp16)[name = string("linear_190_cast_fp16")]; + string x_389_mode_0 = const()[name = string("x_389_mode_0"), val = string("EXACT")]; + tensor x_389_cast_fp16 = gelu(mode = x_389_mode_0, x = linear_190_cast_fp16)[name = string("x_389_cast_fp16")]; + tensor var_3153_to_fp16 = const()[name = string("op_3153_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(616545280)))]; + tensor var_3154_to_fp16 = const()[name = string("op_3154_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(629652544)))]; + tensor linear_191_cast_fp16 = linear(bias = var_3154_to_fp16, weight = var_3153_to_fp16, x = x_389_cast_fp16)[name = string("linear_191_cast_fp16")]; + tensor x_cast_fp16 = add(x = x_385_cast_fp16, y = linear_191_cast_fp16)[name = string("x_cast_fp16")]; + tensor var_3167_axes_0 = const()[name = string("op_3167_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = string("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(629655168)))]; + tensor ln_post_bias_to_fp16 = const()[name = string("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(629657792)))]; + fp16 var_3158_to_fp16 = const()[name = string("op_3158_to_fp16"), val = fp16(0x1.5p-17)]; + tensor output = layer_norm(axes = var_3167_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3158_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = string("op_3167_cast_fp16")]; + } -> (output); +} \ No newline at end of file