diff --git "a/ggml-large-v3-turbo-encoder.mlmodelc/model0/model.mil" "b/ggml-large-v3-turbo-encoder.mlmodelc/model0/model.mil" new file mode 100644--- /dev/null +++ "b/ggml-large-v3-turbo-encoder.mlmodelc/model0/model.mil" @@ -0,0 +1,786 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3400.43.1"}, {"coremlc-version", "3400.58.2"}})] +{ + func main(tensor logmel_data) { + string var_84_pad_type_0 = const()[name = string("op_84_pad_type_0"), val = string("custom")]; + tensor var_84_pad_0 = const()[name = string("op_84_pad_0"), val = tensor([1, 1])]; + tensor var_84_strides_0 = const()[name = string("op_84_strides_0"), val = tensor([1])]; + tensor var_84_dilations_0 = const()[name = string("op_84_dilations_0"), val = tensor([1])]; + int32 var_84_groups_0 = const()[name = string("op_84_groups_0"), val = int32(1)]; + tensor weight_3_to_fp16 = const()[name = string("weight_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(64)))]; + tensor bias_3_to_fp16 = const()[name = string("bias_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(983168)))]; + tensor var_84_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_84_dilations_0, groups = var_84_groups_0, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_84_strides_0, weight = weight_3_to_fp16, x = logmel_data)[name = string("op_84_cast_fp16")]; + string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; + tensor input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_84_cast_fp16)[name = string("input_1_cast_fp16")]; + string var_102_pad_type_0 = const()[name = string("op_102_pad_type_0"), val = string("custom")]; + tensor var_102_pad_0 = const()[name = string("op_102_pad_0"), val = tensor([1, 1])]; + tensor var_102_strides_0 = const()[name = string("op_102_strides_0"), val = tensor([2])]; + tensor var_102_dilations_0 = const()[name = string("op_102_dilations_0"), val = tensor([1])]; + int32 var_102_groups_0 = const()[name = string("op_102_groups_0"), val = int32(1)]; + tensor weight_7_to_fp16 = const()[name = string("weight_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(985792)))]; + tensor bias_7_to_fp16 = const()[name = string("bias_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(10816256)))]; + tensor var_102_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_102_dilations_0, groups = var_102_groups_0, pad = var_102_pad_0, pad_type = var_102_pad_type_0, strides = var_102_strides_0, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = string("op_102_cast_fp16")]; + string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; + tensor x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_102_cast_fp16)[name = string("x_3_cast_fp16")]; + tensor var_108 = const()[name = string("op_108"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(10818880)))]; + tensor x_5_cast_fp16 = transpose(perm = var_108, x = x_3_cast_fp16)[name = string("transpose_208")]; + tensor var_111_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_111_cast_fp16")]; + tensor var_139_axes_0 = const()[name = string("op_139_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(14658944)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(14661568)))]; + fp16 var_129_to_fp16 = const()[name = string("op_129_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_139_cast_fp16 = layer_norm(axes = var_139_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_129_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_111_cast_fp16)[name = string("op_139_cast_fp16")]; + tensor var_150_to_fp16 = const()[name = string("op_150_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(14664192)))]; + tensor var_151_to_fp16 = const()[name = string("op_151_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(17941056)))]; + tensor linear_0_cast_fp16 = linear(bias = var_151_to_fp16, weight = var_150_to_fp16, x = var_139_cast_fp16)[name = string("linear_0_cast_fp16")]; + tensor var_154_to_fp16 = const()[name = string("op_154_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(17943680)))]; + tensor linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(21220544)))]; + tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_154_to_fp16, x = var_139_cast_fp16)[name = string("linear_1_cast_fp16")]; + tensor var_158_to_fp16 = const()[name = string("op_158_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(21223168)))]; + tensor var_159_to_fp16 = const()[name = string("op_159_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(24500032)))]; + tensor linear_2_cast_fp16 = linear(bias = var_159_to_fp16, weight = var_158_to_fp16, x = var_139_cast_fp16)[name = string("linear_2_cast_fp16")]; + tensor var_163 = const()[name = string("op_163"), val = tensor([1, 1500, 20, -1])]; + tensor var_164_cast_fp16 = reshape(shape = var_163, x = linear_0_cast_fp16)[name = string("op_164_cast_fp16")]; + tensor var_169 = const()[name = string("op_169"), val = tensor([1, 1500, 20, -1])]; + tensor var_170_cast_fp16 = reshape(shape = var_169, x = linear_1_cast_fp16)[name = string("op_170_cast_fp16")]; + tensor var_175 = const()[name = string("op_175"), val = tensor([1, 1500, 20, -1])]; + tensor var_176_cast_fp16 = reshape(shape = var_175, x = linear_2_cast_fp16)[name = string("op_176_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_176_cast_fp16)[name = string("transpose_205")]; + tensor transpose_97 = transpose(perm = transpose_97_perm_0, x = var_170_cast_fp16)[name = string("transpose_206")]; + tensor transpose_96 = transpose(perm = transpose_96_perm_0, x = var_164_cast_fp16)[name = string("transpose_207")]; + tensor a_1_cast_fp16 = scaled_dot_product_attention(key = transpose_97, query = transpose_96, value = transpose_98)[name = string("a_1_cast_fp16")]; + tensor var_180 = const()[name = string("op_180"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = string("concat_0"), val = tensor([1, 1500, 1280])]; + tensor var_181_cast_fp16 = transpose(perm = var_180, x = a_1_cast_fp16)[name = string("transpose_204")]; + tensor x_11_cast_fp16 = reshape(shape = concat_0, x = var_181_cast_fp16)[name = string("x_11_cast_fp16")]; + tensor var_185_to_fp16 = const()[name = string("op_185_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(24502656)))]; + tensor var_186_to_fp16 = const()[name = string("op_186_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(27779520)))]; + tensor linear_3_cast_fp16 = linear(bias = var_186_to_fp16, weight = var_185_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")]; + tensor x_13_cast_fp16 = add(x = var_111_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; + tensor var_193_axes_0 = const()[name = string("op_193_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(27782144)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(27784768)))]; + tensor var_193_cast_fp16 = layer_norm(axes = var_193_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_129_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_193_cast_fp16")]; + tensor var_202_to_fp16 = const()[name = string("op_202_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(27787392)))]; + tensor var_203_to_fp16 = const()[name = string("op_203_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(40894656)))]; + tensor linear_4_cast_fp16 = linear(bias = var_203_to_fp16, weight = var_202_to_fp16, x = var_193_cast_fp16)[name = string("linear_4_cast_fp16")]; + string x_17_mode_0 = const()[name = string("x_17_mode_0"), val = string("EXACT")]; + tensor x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; + tensor var_208_to_fp16 = const()[name = string("op_208_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(40904960)))]; + tensor var_209_to_fp16 = const()[name = string("op_209_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(54012224)))]; + tensor linear_5_cast_fp16 = linear(bias = var_209_to_fp16, weight = var_208_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")]; + tensor x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")]; + tensor var_234_axes_0 = const()[name = string("op_234_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(54014848)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(54017472)))]; + fp16 var_224_to_fp16 = const()[name = string("op_224_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_234_cast_fp16 = layer_norm(axes = var_234_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_224_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_234_cast_fp16")]; + tensor var_245_to_fp16 = const()[name = string("op_245_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(54020096)))]; + tensor var_246_to_fp16 = const()[name = string("op_246_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(57296960)))]; + tensor linear_6_cast_fp16 = linear(bias = var_246_to_fp16, weight = var_245_to_fp16, x = var_234_cast_fp16)[name = string("linear_6_cast_fp16")]; + tensor var_249_to_fp16 = const()[name = string("op_249_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(57299584)))]; + tensor linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_249_to_fp16, x = var_234_cast_fp16)[name = string("linear_7_cast_fp16")]; + tensor var_253_to_fp16 = const()[name = string("op_253_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(60576448)))]; + tensor var_254_to_fp16 = const()[name = string("op_254_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(63853312)))]; + tensor linear_8_cast_fp16 = linear(bias = var_254_to_fp16, weight = var_253_to_fp16, x = var_234_cast_fp16)[name = string("linear_8_cast_fp16")]; + tensor var_258 = const()[name = string("op_258"), val = tensor([1, 1500, 20, -1])]; + tensor var_259_cast_fp16 = reshape(shape = var_258, x = linear_6_cast_fp16)[name = string("op_259_cast_fp16")]; + tensor var_264 = const()[name = string("op_264"), val = tensor([1, 1500, 20, -1])]; + tensor var_265_cast_fp16 = reshape(shape = var_264, x = linear_7_cast_fp16)[name = string("op_265_cast_fp16")]; + tensor var_270 = const()[name = string("op_270"), val = tensor([1, 1500, 20, -1])]; + tensor var_271_cast_fp16 = reshape(shape = var_270, x = linear_8_cast_fp16)[name = string("op_271_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_271_cast_fp16)[name = string("transpose_201")]; + tensor transpose_100 = transpose(perm = transpose_100_perm_0, x = var_265_cast_fp16)[name = string("transpose_202")]; + tensor transpose_99 = transpose(perm = transpose_99_perm_0, x = var_259_cast_fp16)[name = string("transpose_203")]; + tensor a_3_cast_fp16 = scaled_dot_product_attention(key = transpose_100, query = transpose_99, value = transpose_101)[name = string("a_3_cast_fp16")]; + tensor var_275 = const()[name = string("op_275"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = string("concat_1"), val = tensor([1, 1500, 1280])]; + tensor var_276_cast_fp16 = transpose(perm = var_275, x = a_3_cast_fp16)[name = string("transpose_200")]; + tensor x_23_cast_fp16 = reshape(shape = concat_1, x = var_276_cast_fp16)[name = string("x_23_cast_fp16")]; + tensor var_280_to_fp16 = const()[name = string("op_280_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(63855936)))]; + tensor var_281_to_fp16 = const()[name = string("op_281_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(67132800)))]; + tensor linear_9_cast_fp16 = linear(bias = var_281_to_fp16, weight = var_280_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")]; + tensor x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = string("x_25_cast_fp16")]; + tensor var_288_axes_0 = const()[name = string("op_288_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(67135424)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(67138048)))]; + tensor var_288_cast_fp16 = layer_norm(axes = var_288_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_224_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_288_cast_fp16")]; + tensor var_297_to_fp16 = const()[name = string("op_297_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(67140672)))]; + tensor var_298_to_fp16 = const()[name = string("op_298_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(80247936)))]; + tensor linear_10_cast_fp16 = linear(bias = var_298_to_fp16, weight = var_297_to_fp16, x = var_288_cast_fp16)[name = string("linear_10_cast_fp16")]; + string x_29_mode_0 = const()[name = string("x_29_mode_0"), val = string("EXACT")]; + tensor x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; + tensor var_303_to_fp16 = const()[name = string("op_303_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(80258240)))]; + tensor var_304_to_fp16 = const()[name = string("op_304_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(93365504)))]; + tensor linear_11_cast_fp16 = linear(bias = var_304_to_fp16, weight = var_303_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")]; + tensor x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")]; + tensor var_329_axes_0 = const()[name = string("op_329_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(93368128)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(93370752)))]; + fp16 var_319_to_fp16 = const()[name = string("op_319_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_329_cast_fp16 = layer_norm(axes = var_329_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_319_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_329_cast_fp16")]; + tensor var_340_to_fp16 = const()[name = string("op_340_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(93373376)))]; + tensor var_341_to_fp16 = const()[name = string("op_341_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(96650240)))]; + tensor linear_12_cast_fp16 = linear(bias = var_341_to_fp16, weight = var_340_to_fp16, x = var_329_cast_fp16)[name = string("linear_12_cast_fp16")]; + tensor var_344_to_fp16 = const()[name = string("op_344_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(96652864)))]; + tensor linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_344_to_fp16, x = var_329_cast_fp16)[name = string("linear_13_cast_fp16")]; + tensor var_348_to_fp16 = const()[name = string("op_348_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(99929728)))]; + tensor var_349_to_fp16 = const()[name = string("op_349_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(103206592)))]; + tensor linear_14_cast_fp16 = linear(bias = var_349_to_fp16, weight = var_348_to_fp16, x = var_329_cast_fp16)[name = string("linear_14_cast_fp16")]; + tensor var_353 = const()[name = string("op_353"), val = tensor([1, 1500, 20, -1])]; + tensor var_354_cast_fp16 = reshape(shape = var_353, x = linear_12_cast_fp16)[name = string("op_354_cast_fp16")]; + tensor var_359 = const()[name = string("op_359"), val = tensor([1, 1500, 20, -1])]; + tensor var_360_cast_fp16 = reshape(shape = var_359, x = linear_13_cast_fp16)[name = string("op_360_cast_fp16")]; + tensor var_365 = const()[name = string("op_365"), val = tensor([1, 1500, 20, -1])]; + tensor var_366_cast_fp16 = reshape(shape = var_365, x = linear_14_cast_fp16)[name = string("op_366_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_366_cast_fp16)[name = string("transpose_197")]; + tensor transpose_103 = transpose(perm = transpose_103_perm_0, x = var_360_cast_fp16)[name = string("transpose_198")]; + tensor transpose_102 = transpose(perm = transpose_102_perm_0, x = var_354_cast_fp16)[name = string("transpose_199")]; + tensor a_5_cast_fp16 = scaled_dot_product_attention(key = transpose_103, query = transpose_102, value = transpose_104)[name = string("a_5_cast_fp16")]; + tensor var_370 = const()[name = string("op_370"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = string("concat_2"), val = tensor([1, 1500, 1280])]; + tensor var_371_cast_fp16 = transpose(perm = var_370, x = a_5_cast_fp16)[name = string("transpose_196")]; + tensor x_35_cast_fp16 = reshape(shape = concat_2, x = var_371_cast_fp16)[name = string("x_35_cast_fp16")]; + tensor var_375_to_fp16 = const()[name = string("op_375_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(103209216)))]; + tensor var_376_to_fp16 = const()[name = string("op_376_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(106486080)))]; + tensor linear_15_cast_fp16 = linear(bias = var_376_to_fp16, weight = var_375_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")]; + tensor x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = string("x_37_cast_fp16")]; + tensor var_383_axes_0 = const()[name = string("op_383_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(106488704)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(106491328)))]; + tensor var_383_cast_fp16 = layer_norm(axes = var_383_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_319_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_383_cast_fp16")]; + tensor var_392_to_fp16 = const()[name = string("op_392_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(106493952)))]; + tensor var_393_to_fp16 = const()[name = string("op_393_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(119601216)))]; + tensor linear_16_cast_fp16 = linear(bias = var_393_to_fp16, weight = var_392_to_fp16, x = var_383_cast_fp16)[name = string("linear_16_cast_fp16")]; + string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("EXACT")]; + tensor x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; + tensor var_398_to_fp16 = const()[name = string("op_398_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(119611520)))]; + tensor var_399_to_fp16 = const()[name = string("op_399_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(132718784)))]; + tensor linear_17_cast_fp16 = linear(bias = var_399_to_fp16, weight = var_398_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")]; + tensor x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")]; + tensor var_424_axes_0 = const()[name = string("op_424_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(132721408)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(132724032)))]; + fp16 var_414_to_fp16 = const()[name = string("op_414_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_424_cast_fp16 = layer_norm(axes = var_424_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_414_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_424_cast_fp16")]; + tensor var_435_to_fp16 = const()[name = string("op_435_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(132726656)))]; + tensor var_436_to_fp16 = const()[name = string("op_436_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(136003520)))]; + tensor linear_18_cast_fp16 = linear(bias = var_436_to_fp16, weight = var_435_to_fp16, x = var_424_cast_fp16)[name = string("linear_18_cast_fp16")]; + tensor var_439_to_fp16 = const()[name = string("op_439_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(136006144)))]; + tensor linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_439_to_fp16, x = var_424_cast_fp16)[name = string("linear_19_cast_fp16")]; + tensor var_443_to_fp16 = const()[name = string("op_443_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(139283008)))]; + tensor var_444_to_fp16 = const()[name = string("op_444_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(142559872)))]; + tensor linear_20_cast_fp16 = linear(bias = var_444_to_fp16, weight = var_443_to_fp16, x = var_424_cast_fp16)[name = string("linear_20_cast_fp16")]; + tensor var_448 = const()[name = string("op_448"), val = tensor([1, 1500, 20, -1])]; + tensor var_449_cast_fp16 = reshape(shape = var_448, x = linear_18_cast_fp16)[name = string("op_449_cast_fp16")]; + tensor var_454 = const()[name = string("op_454"), val = tensor([1, 1500, 20, -1])]; + tensor var_455_cast_fp16 = reshape(shape = var_454, x = linear_19_cast_fp16)[name = string("op_455_cast_fp16")]; + tensor var_460 = const()[name = string("op_460"), val = tensor([1, 1500, 20, -1])]; + tensor var_461_cast_fp16 = reshape(shape = var_460, x = linear_20_cast_fp16)[name = string("op_461_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_461_cast_fp16)[name = string("transpose_193")]; + tensor transpose_106 = transpose(perm = transpose_106_perm_0, x = var_455_cast_fp16)[name = string("transpose_194")]; + tensor transpose_105 = transpose(perm = transpose_105_perm_0, x = var_449_cast_fp16)[name = string("transpose_195")]; + tensor a_7_cast_fp16 = scaled_dot_product_attention(key = transpose_106, query = transpose_105, value = transpose_107)[name = string("a_7_cast_fp16")]; + tensor var_465 = const()[name = string("op_465"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = string("concat_3"), val = tensor([1, 1500, 1280])]; + tensor var_466_cast_fp16 = transpose(perm = var_465, x = a_7_cast_fp16)[name = string("transpose_192")]; + tensor x_47_cast_fp16 = reshape(shape = concat_3, x = var_466_cast_fp16)[name = string("x_47_cast_fp16")]; + tensor var_470_to_fp16 = const()[name = string("op_470_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(142562496)))]; + tensor var_471_to_fp16 = const()[name = string("op_471_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(145839360)))]; + tensor linear_21_cast_fp16 = linear(bias = var_471_to_fp16, weight = var_470_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")]; + tensor x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = string("x_49_cast_fp16")]; + tensor var_478_axes_0 = const()[name = string("op_478_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(145841984)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(145844608)))]; + tensor var_478_cast_fp16 = layer_norm(axes = var_478_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_414_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_478_cast_fp16")]; + tensor var_487_to_fp16 = const()[name = string("op_487_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(145847232)))]; + tensor var_488_to_fp16 = const()[name = string("op_488_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(158954496)))]; + tensor linear_22_cast_fp16 = linear(bias = var_488_to_fp16, weight = var_487_to_fp16, x = var_478_cast_fp16)[name = string("linear_22_cast_fp16")]; + string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("EXACT")]; + tensor x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; + tensor var_493_to_fp16 = const()[name = string("op_493_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(158964800)))]; + tensor var_494_to_fp16 = const()[name = string("op_494_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(172072064)))]; + tensor linear_23_cast_fp16 = linear(bias = var_494_to_fp16, weight = var_493_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")]; + tensor x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_55_cast_fp16")]; + tensor var_519_axes_0 = const()[name = string("op_519_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = string("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(172074688)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = string("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(172077312)))]; + fp16 var_509_to_fp16 = const()[name = string("op_509_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_519_cast_fp16 = layer_norm(axes = var_519_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_509_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = string("op_519_cast_fp16")]; + tensor var_530_to_fp16 = const()[name = string("op_530_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(172079936)))]; + tensor var_531_to_fp16 = const()[name = string("op_531_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(175356800)))]; + tensor linear_24_cast_fp16 = linear(bias = var_531_to_fp16, weight = var_530_to_fp16, x = var_519_cast_fp16)[name = string("linear_24_cast_fp16")]; + tensor var_534_to_fp16 = const()[name = string("op_534_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(175359424)))]; + tensor linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_534_to_fp16, x = var_519_cast_fp16)[name = string("linear_25_cast_fp16")]; + tensor var_538_to_fp16 = const()[name = string("op_538_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(178636288)))]; + tensor var_539_to_fp16 = const()[name = string("op_539_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(181913152)))]; + tensor linear_26_cast_fp16 = linear(bias = var_539_to_fp16, weight = var_538_to_fp16, x = var_519_cast_fp16)[name = string("linear_26_cast_fp16")]; + tensor var_543 = const()[name = string("op_543"), val = tensor([1, 1500, 20, -1])]; + tensor var_544_cast_fp16 = reshape(shape = var_543, x = linear_24_cast_fp16)[name = string("op_544_cast_fp16")]; + tensor var_549 = const()[name = string("op_549"), val = tensor([1, 1500, 20, -1])]; + tensor var_550_cast_fp16 = reshape(shape = var_549, x = linear_25_cast_fp16)[name = string("op_550_cast_fp16")]; + tensor var_555 = const()[name = string("op_555"), val = tensor([1, 1500, 20, -1])]; + tensor var_556_cast_fp16 = reshape(shape = var_555, x = linear_26_cast_fp16)[name = string("op_556_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_556_cast_fp16)[name = string("transpose_189")]; + tensor transpose_109 = transpose(perm = transpose_109_perm_0, x = var_550_cast_fp16)[name = string("transpose_190")]; + tensor transpose_108 = transpose(perm = transpose_108_perm_0, x = var_544_cast_fp16)[name = string("transpose_191")]; + tensor a_9_cast_fp16 = scaled_dot_product_attention(key = transpose_109, query = transpose_108, value = transpose_110)[name = string("a_9_cast_fp16")]; + tensor var_560 = const()[name = string("op_560"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = string("concat_4"), val = tensor([1, 1500, 1280])]; + tensor var_561_cast_fp16 = transpose(perm = var_560, x = a_9_cast_fp16)[name = string("transpose_188")]; + tensor x_59_cast_fp16 = reshape(shape = concat_4, x = var_561_cast_fp16)[name = string("x_59_cast_fp16")]; + tensor var_565_to_fp16 = const()[name = string("op_565_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(181915776)))]; + tensor var_566_to_fp16 = const()[name = string("op_566_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(185192640)))]; + tensor linear_27_cast_fp16 = linear(bias = var_566_to_fp16, weight = var_565_to_fp16, x = x_59_cast_fp16)[name = string("linear_27_cast_fp16")]; + tensor x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = string("x_61_cast_fp16")]; + tensor var_573_axes_0 = const()[name = string("op_573_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = string("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(185195264)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = string("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(185197888)))]; + tensor var_573_cast_fp16 = layer_norm(axes = var_573_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_509_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = string("op_573_cast_fp16")]; + tensor var_582_to_fp16 = const()[name = string("op_582_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(185200512)))]; + tensor var_583_to_fp16 = const()[name = string("op_583_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(198307776)))]; + tensor linear_28_cast_fp16 = linear(bias = var_583_to_fp16, weight = var_582_to_fp16, x = var_573_cast_fp16)[name = string("linear_28_cast_fp16")]; + string x_65_mode_0 = const()[name = string("x_65_mode_0"), val = string("EXACT")]; + tensor x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = string("x_65_cast_fp16")]; + tensor var_588_to_fp16 = const()[name = string("op_588_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(198318080)))]; + tensor var_589_to_fp16 = const()[name = string("op_589_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(211425344)))]; + tensor linear_29_cast_fp16 = linear(bias = var_589_to_fp16, weight = var_588_to_fp16, x = x_65_cast_fp16)[name = string("linear_29_cast_fp16")]; + tensor x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = string("x_67_cast_fp16")]; + tensor var_614_axes_0 = const()[name = string("op_614_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = string("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(211427968)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = string("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(211430592)))]; + fp16 var_604_to_fp16 = const()[name = string("op_604_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_614_cast_fp16 = layer_norm(axes = var_614_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_604_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = string("op_614_cast_fp16")]; + tensor var_625_to_fp16 = const()[name = string("op_625_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(211433216)))]; + tensor var_626_to_fp16 = const()[name = string("op_626_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(214710080)))]; + tensor linear_30_cast_fp16 = linear(bias = var_626_to_fp16, weight = var_625_to_fp16, x = var_614_cast_fp16)[name = string("linear_30_cast_fp16")]; + tensor var_629_to_fp16 = const()[name = string("op_629_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(214712704)))]; + tensor linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_629_to_fp16, x = var_614_cast_fp16)[name = string("linear_31_cast_fp16")]; + tensor var_633_to_fp16 = const()[name = string("op_633_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(217989568)))]; + tensor var_634_to_fp16 = const()[name = string("op_634_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(221266432)))]; + tensor linear_32_cast_fp16 = linear(bias = var_634_to_fp16, weight = var_633_to_fp16, x = var_614_cast_fp16)[name = string("linear_32_cast_fp16")]; + tensor var_638 = const()[name = string("op_638"), val = tensor([1, 1500, 20, -1])]; + tensor var_639_cast_fp16 = reshape(shape = var_638, x = linear_30_cast_fp16)[name = string("op_639_cast_fp16")]; + tensor var_644 = const()[name = string("op_644"), val = tensor([1, 1500, 20, -1])]; + tensor var_645_cast_fp16 = reshape(shape = var_644, x = linear_31_cast_fp16)[name = string("op_645_cast_fp16")]; + tensor var_650 = const()[name = string("op_650"), val = tensor([1, 1500, 20, -1])]; + tensor var_651_cast_fp16 = reshape(shape = var_650, x = linear_32_cast_fp16)[name = string("op_651_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_651_cast_fp16)[name = string("transpose_185")]; + tensor transpose_112 = transpose(perm = transpose_112_perm_0, x = var_645_cast_fp16)[name = string("transpose_186")]; + tensor transpose_111 = transpose(perm = transpose_111_perm_0, x = var_639_cast_fp16)[name = string("transpose_187")]; + tensor a_11_cast_fp16 = scaled_dot_product_attention(key = transpose_112, query = transpose_111, value = transpose_113)[name = string("a_11_cast_fp16")]; + tensor var_655 = const()[name = string("op_655"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = string("concat_5"), val = tensor([1, 1500, 1280])]; + tensor var_656_cast_fp16 = transpose(perm = var_655, x = a_11_cast_fp16)[name = string("transpose_184")]; + tensor x_71_cast_fp16 = reshape(shape = concat_5, x = var_656_cast_fp16)[name = string("x_71_cast_fp16")]; + tensor var_660_to_fp16 = const()[name = string("op_660_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(221269056)))]; + tensor var_661_to_fp16 = const()[name = string("op_661_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(224545920)))]; + tensor linear_33_cast_fp16 = linear(bias = var_661_to_fp16, weight = var_660_to_fp16, x = x_71_cast_fp16)[name = string("linear_33_cast_fp16")]; + tensor x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_73_cast_fp16")]; + tensor var_668_axes_0 = const()[name = string("op_668_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = string("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(224548544)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = string("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(224551168)))]; + tensor var_668_cast_fp16 = layer_norm(axes = var_668_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_604_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = string("op_668_cast_fp16")]; + tensor var_677_to_fp16 = const()[name = string("op_677_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(224553792)))]; + tensor var_678_to_fp16 = const()[name = string("op_678_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(237661056)))]; + tensor linear_34_cast_fp16 = linear(bias = var_678_to_fp16, weight = var_677_to_fp16, x = var_668_cast_fp16)[name = string("linear_34_cast_fp16")]; + string x_77_mode_0 = const()[name = string("x_77_mode_0"), val = string("EXACT")]; + tensor x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = string("x_77_cast_fp16")]; + tensor var_683_to_fp16 = const()[name = string("op_683_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(237671360)))]; + tensor var_684_to_fp16 = const()[name = string("op_684_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(250778624)))]; + tensor linear_35_cast_fp16 = linear(bias = var_684_to_fp16, weight = var_683_to_fp16, x = x_77_cast_fp16)[name = string("linear_35_cast_fp16")]; + tensor x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = string("x_79_cast_fp16")]; + tensor var_709_axes_0 = const()[name = string("op_709_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = string("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(250781248)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = string("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(250783872)))]; + fp16 var_699_to_fp16 = const()[name = string("op_699_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_709_cast_fp16 = layer_norm(axes = var_709_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_699_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = string("op_709_cast_fp16")]; + tensor var_720_to_fp16 = const()[name = string("op_720_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(250786496)))]; + tensor var_721_to_fp16 = const()[name = string("op_721_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(254063360)))]; + tensor linear_36_cast_fp16 = linear(bias = var_721_to_fp16, weight = var_720_to_fp16, x = var_709_cast_fp16)[name = string("linear_36_cast_fp16")]; + tensor var_724_to_fp16 = const()[name = string("op_724_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(254065984)))]; + tensor linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_724_to_fp16, x = var_709_cast_fp16)[name = string("linear_37_cast_fp16")]; + tensor var_728_to_fp16 = const()[name = string("op_728_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(257342848)))]; + tensor var_729_to_fp16 = const()[name = string("op_729_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(260619712)))]; + tensor linear_38_cast_fp16 = linear(bias = var_729_to_fp16, weight = var_728_to_fp16, x = var_709_cast_fp16)[name = string("linear_38_cast_fp16")]; + tensor var_733 = const()[name = string("op_733"), val = tensor([1, 1500, 20, -1])]; + tensor var_734_cast_fp16 = reshape(shape = var_733, x = linear_36_cast_fp16)[name = string("op_734_cast_fp16")]; + tensor var_739 = const()[name = string("op_739"), val = tensor([1, 1500, 20, -1])]; + tensor var_740_cast_fp16 = reshape(shape = var_739, x = linear_37_cast_fp16)[name = string("op_740_cast_fp16")]; + tensor var_745 = const()[name = string("op_745"), val = tensor([1, 1500, 20, -1])]; + tensor var_746_cast_fp16 = reshape(shape = var_745, x = linear_38_cast_fp16)[name = string("op_746_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_746_cast_fp16)[name = string("transpose_181")]; + tensor transpose_115 = transpose(perm = transpose_115_perm_0, x = var_740_cast_fp16)[name = string("transpose_182")]; + tensor transpose_114 = transpose(perm = transpose_114_perm_0, x = var_734_cast_fp16)[name = string("transpose_183")]; + tensor a_13_cast_fp16 = scaled_dot_product_attention(key = transpose_115, query = transpose_114, value = transpose_116)[name = string("a_13_cast_fp16")]; + tensor var_750 = const()[name = string("op_750"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = string("concat_6"), val = tensor([1, 1500, 1280])]; + tensor var_751_cast_fp16 = transpose(perm = var_750, x = a_13_cast_fp16)[name = string("transpose_180")]; + tensor x_83_cast_fp16 = reshape(shape = concat_6, x = var_751_cast_fp16)[name = string("x_83_cast_fp16")]; + tensor var_755_to_fp16 = const()[name = string("op_755_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(260622336)))]; + tensor var_756_to_fp16 = const()[name = string("op_756_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(263899200)))]; + tensor linear_39_cast_fp16 = linear(bias = var_756_to_fp16, weight = var_755_to_fp16, x = x_83_cast_fp16)[name = string("linear_39_cast_fp16")]; + tensor x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = string("x_85_cast_fp16")]; + tensor var_763_axes_0 = const()[name = string("op_763_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = string("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(263901824)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = string("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(263904448)))]; + tensor var_763_cast_fp16 = layer_norm(axes = var_763_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_699_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = string("op_763_cast_fp16")]; + tensor var_772_to_fp16 = const()[name = string("op_772_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(263907072)))]; + tensor var_773_to_fp16 = const()[name = string("op_773_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(277014336)))]; + tensor linear_40_cast_fp16 = linear(bias = var_773_to_fp16, weight = var_772_to_fp16, x = var_763_cast_fp16)[name = string("linear_40_cast_fp16")]; + string x_89_mode_0 = const()[name = string("x_89_mode_0"), val = string("EXACT")]; + tensor x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = string("x_89_cast_fp16")]; + tensor var_778_to_fp16 = const()[name = string("op_778_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(277024640)))]; + tensor var_779_to_fp16 = const()[name = string("op_779_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(290131904)))]; + tensor linear_41_cast_fp16 = linear(bias = var_779_to_fp16, weight = var_778_to_fp16, x = x_89_cast_fp16)[name = string("linear_41_cast_fp16")]; + tensor x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = string("x_91_cast_fp16")]; + tensor var_804_axes_0 = const()[name = string("op_804_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = string("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(290134528)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = string("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(290137152)))]; + fp16 var_794_to_fp16 = const()[name = string("op_794_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_804_cast_fp16 = layer_norm(axes = var_804_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_794_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = string("op_804_cast_fp16")]; + tensor var_815_to_fp16 = const()[name = string("op_815_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(290139776)))]; + tensor var_816_to_fp16 = const()[name = string("op_816_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(293416640)))]; + tensor linear_42_cast_fp16 = linear(bias = var_816_to_fp16, weight = var_815_to_fp16, x = var_804_cast_fp16)[name = string("linear_42_cast_fp16")]; + tensor var_819_to_fp16 = const()[name = string("op_819_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(293419264)))]; + tensor linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_819_to_fp16, x = var_804_cast_fp16)[name = string("linear_43_cast_fp16")]; + tensor var_823_to_fp16 = const()[name = string("op_823_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(296696128)))]; + tensor var_824_to_fp16 = const()[name = string("op_824_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(299972992)))]; + tensor linear_44_cast_fp16 = linear(bias = var_824_to_fp16, weight = var_823_to_fp16, x = var_804_cast_fp16)[name = string("linear_44_cast_fp16")]; + tensor var_828 = const()[name = string("op_828"), val = tensor([1, 1500, 20, -1])]; + tensor var_829_cast_fp16 = reshape(shape = var_828, x = linear_42_cast_fp16)[name = string("op_829_cast_fp16")]; + tensor var_834 = const()[name = string("op_834"), val = tensor([1, 1500, 20, -1])]; + tensor var_835_cast_fp16 = reshape(shape = var_834, x = linear_43_cast_fp16)[name = string("op_835_cast_fp16")]; + tensor var_840 = const()[name = string("op_840"), val = tensor([1, 1500, 20, -1])]; + tensor var_841_cast_fp16 = reshape(shape = var_840, x = linear_44_cast_fp16)[name = string("op_841_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_841_cast_fp16)[name = string("transpose_177")]; + tensor transpose_118 = transpose(perm = transpose_118_perm_0, x = var_835_cast_fp16)[name = string("transpose_178")]; + tensor transpose_117 = transpose(perm = transpose_117_perm_0, x = var_829_cast_fp16)[name = string("transpose_179")]; + tensor a_15_cast_fp16 = scaled_dot_product_attention(key = transpose_118, query = transpose_117, value = transpose_119)[name = string("a_15_cast_fp16")]; + tensor var_845 = const()[name = string("op_845"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = string("concat_7"), val = tensor([1, 1500, 1280])]; + tensor var_846_cast_fp16 = transpose(perm = var_845, x = a_15_cast_fp16)[name = string("transpose_176")]; + tensor x_95_cast_fp16 = reshape(shape = concat_7, x = var_846_cast_fp16)[name = string("x_95_cast_fp16")]; + tensor var_850_to_fp16 = const()[name = string("op_850_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(299975616)))]; + tensor var_851_to_fp16 = const()[name = string("op_851_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(303252480)))]; + tensor linear_45_cast_fp16 = linear(bias = var_851_to_fp16, weight = var_850_to_fp16, x = x_95_cast_fp16)[name = string("linear_45_cast_fp16")]; + tensor x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = string("x_97_cast_fp16")]; + tensor var_858_axes_0 = const()[name = string("op_858_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = string("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(303255104)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = string("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(303257728)))]; + tensor var_858_cast_fp16 = layer_norm(axes = var_858_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_794_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = string("op_858_cast_fp16")]; + tensor var_867_to_fp16 = const()[name = string("op_867_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(303260352)))]; + tensor var_868_to_fp16 = const()[name = string("op_868_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(316367616)))]; + tensor linear_46_cast_fp16 = linear(bias = var_868_to_fp16, weight = var_867_to_fp16, x = var_858_cast_fp16)[name = string("linear_46_cast_fp16")]; + string x_101_mode_0 = const()[name = string("x_101_mode_0"), val = string("EXACT")]; + tensor x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = string("x_101_cast_fp16")]; + tensor var_873_to_fp16 = const()[name = string("op_873_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(316377920)))]; + tensor var_874_to_fp16 = const()[name = string("op_874_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(329485184)))]; + tensor linear_47_cast_fp16 = linear(bias = var_874_to_fp16, weight = var_873_to_fp16, x = x_101_cast_fp16)[name = string("linear_47_cast_fp16")]; + tensor x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = string("x_103_cast_fp16")]; + tensor var_899_axes_0 = const()[name = string("op_899_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = string("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(329487808)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = string("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(329490432)))]; + fp16 var_889_to_fp16 = const()[name = string("op_889_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_899_cast_fp16 = layer_norm(axes = var_899_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_889_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = string("op_899_cast_fp16")]; + tensor var_910_to_fp16 = const()[name = string("op_910_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(329493056)))]; + tensor var_911_to_fp16 = const()[name = string("op_911_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(332769920)))]; + tensor linear_48_cast_fp16 = linear(bias = var_911_to_fp16, weight = var_910_to_fp16, x = var_899_cast_fp16)[name = string("linear_48_cast_fp16")]; + tensor var_914_to_fp16 = const()[name = string("op_914_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(332772544)))]; + tensor linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_914_to_fp16, x = var_899_cast_fp16)[name = string("linear_49_cast_fp16")]; + tensor var_918_to_fp16 = const()[name = string("op_918_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(336049408)))]; + tensor var_919_to_fp16 = const()[name = string("op_919_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(339326272)))]; + tensor linear_50_cast_fp16 = linear(bias = var_919_to_fp16, weight = var_918_to_fp16, x = var_899_cast_fp16)[name = string("linear_50_cast_fp16")]; + tensor var_923 = const()[name = string("op_923"), val = tensor([1, 1500, 20, -1])]; + tensor var_924_cast_fp16 = reshape(shape = var_923, x = linear_48_cast_fp16)[name = string("op_924_cast_fp16")]; + tensor var_929 = const()[name = string("op_929"), val = tensor([1, 1500, 20, -1])]; + tensor var_930_cast_fp16 = reshape(shape = var_929, x = linear_49_cast_fp16)[name = string("op_930_cast_fp16")]; + tensor var_935 = const()[name = string("op_935"), val = tensor([1, 1500, 20, -1])]; + tensor var_936_cast_fp16 = reshape(shape = var_935, x = linear_50_cast_fp16)[name = string("op_936_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_936_cast_fp16)[name = string("transpose_173")]; + tensor transpose_121 = transpose(perm = transpose_121_perm_0, x = var_930_cast_fp16)[name = string("transpose_174")]; + tensor transpose_120 = transpose(perm = transpose_120_perm_0, x = var_924_cast_fp16)[name = string("transpose_175")]; + tensor a_17_cast_fp16 = scaled_dot_product_attention(key = transpose_121, query = transpose_120, value = transpose_122)[name = string("a_17_cast_fp16")]; + tensor var_940 = const()[name = string("op_940"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = string("concat_8"), val = tensor([1, 1500, 1280])]; + tensor var_941_cast_fp16 = transpose(perm = var_940, x = a_17_cast_fp16)[name = string("transpose_172")]; + tensor x_107_cast_fp16 = reshape(shape = concat_8, x = var_941_cast_fp16)[name = string("x_107_cast_fp16")]; + tensor var_945_to_fp16 = const()[name = string("op_945_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(339328896)))]; + tensor var_946_to_fp16 = const()[name = string("op_946_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(342605760)))]; + tensor linear_51_cast_fp16 = linear(bias = var_946_to_fp16, weight = var_945_to_fp16, x = x_107_cast_fp16)[name = string("linear_51_cast_fp16")]; + tensor x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = string("x_109_cast_fp16")]; + tensor var_953_axes_0 = const()[name = string("op_953_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = string("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(342608384)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = string("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(342611008)))]; + tensor var_953_cast_fp16 = layer_norm(axes = var_953_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_889_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = string("op_953_cast_fp16")]; + tensor var_962_to_fp16 = const()[name = string("op_962_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(342613632)))]; + tensor var_963_to_fp16 = const()[name = string("op_963_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(355720896)))]; + tensor linear_52_cast_fp16 = linear(bias = var_963_to_fp16, weight = var_962_to_fp16, x = var_953_cast_fp16)[name = string("linear_52_cast_fp16")]; + string x_113_mode_0 = const()[name = string("x_113_mode_0"), val = string("EXACT")]; + tensor x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = string("x_113_cast_fp16")]; + tensor var_968_to_fp16 = const()[name = string("op_968_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(355731200)))]; + tensor var_969_to_fp16 = const()[name = string("op_969_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(368838464)))]; + tensor linear_53_cast_fp16 = linear(bias = var_969_to_fp16, weight = var_968_to_fp16, x = x_113_cast_fp16)[name = string("linear_53_cast_fp16")]; + tensor x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = string("x_115_cast_fp16")]; + tensor var_994_axes_0 = const()[name = string("op_994_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = string("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(368841088)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = string("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(368843712)))]; + fp16 var_984_to_fp16 = const()[name = string("op_984_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_994_cast_fp16 = layer_norm(axes = var_994_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = string("op_994_cast_fp16")]; + tensor var_1005_to_fp16 = const()[name = string("op_1005_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(368846336)))]; + tensor var_1006_to_fp16 = const()[name = string("op_1006_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(372123200)))]; + tensor linear_54_cast_fp16 = linear(bias = var_1006_to_fp16, weight = var_1005_to_fp16, x = var_994_cast_fp16)[name = string("linear_54_cast_fp16")]; + tensor var_1009_to_fp16 = const()[name = string("op_1009_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(372125824)))]; + tensor linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1009_to_fp16, x = var_994_cast_fp16)[name = string("linear_55_cast_fp16")]; + tensor var_1013_to_fp16 = const()[name = string("op_1013_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(375402688)))]; + tensor var_1014_to_fp16 = const()[name = string("op_1014_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(378679552)))]; + tensor linear_56_cast_fp16 = linear(bias = var_1014_to_fp16, weight = var_1013_to_fp16, x = var_994_cast_fp16)[name = string("linear_56_cast_fp16")]; + tensor var_1018 = const()[name = string("op_1018"), val = tensor([1, 1500, 20, -1])]; + tensor var_1019_cast_fp16 = reshape(shape = var_1018, x = linear_54_cast_fp16)[name = string("op_1019_cast_fp16")]; + tensor var_1024 = const()[name = string("op_1024"), val = tensor([1, 1500, 20, -1])]; + tensor var_1025_cast_fp16 = reshape(shape = var_1024, x = linear_55_cast_fp16)[name = string("op_1025_cast_fp16")]; + tensor var_1030 = const()[name = string("op_1030"), val = tensor([1, 1500, 20, -1])]; + tensor var_1031_cast_fp16 = reshape(shape = var_1030, x = linear_56_cast_fp16)[name = string("op_1031_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_1031_cast_fp16)[name = string("transpose_169")]; + tensor transpose_124 = transpose(perm = transpose_124_perm_0, x = var_1025_cast_fp16)[name = string("transpose_170")]; + tensor transpose_123 = transpose(perm = transpose_123_perm_0, x = var_1019_cast_fp16)[name = string("transpose_171")]; + tensor a_19_cast_fp16 = scaled_dot_product_attention(key = transpose_124, query = transpose_123, value = transpose_125)[name = string("a_19_cast_fp16")]; + tensor var_1035 = const()[name = string("op_1035"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = string("concat_9"), val = tensor([1, 1500, 1280])]; + tensor var_1036_cast_fp16 = transpose(perm = var_1035, x = a_19_cast_fp16)[name = string("transpose_168")]; + tensor x_119_cast_fp16 = reshape(shape = concat_9, x = var_1036_cast_fp16)[name = string("x_119_cast_fp16")]; + tensor var_1040_to_fp16 = const()[name = string("op_1040_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(378682176)))]; + tensor var_1041_to_fp16 = const()[name = string("op_1041_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(381959040)))]; + tensor linear_57_cast_fp16 = linear(bias = var_1041_to_fp16, weight = var_1040_to_fp16, x = x_119_cast_fp16)[name = string("linear_57_cast_fp16")]; + tensor x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = string("x_121_cast_fp16")]; + tensor var_1048_axes_0 = const()[name = string("op_1048_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = string("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(381961664)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = string("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(381964288)))]; + tensor var_1048_cast_fp16 = layer_norm(axes = var_1048_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = string("op_1048_cast_fp16")]; + tensor var_1057_to_fp16 = const()[name = string("op_1057_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(381966912)))]; + tensor var_1058_to_fp16 = const()[name = string("op_1058_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(395074176)))]; + tensor linear_58_cast_fp16 = linear(bias = var_1058_to_fp16, weight = var_1057_to_fp16, x = var_1048_cast_fp16)[name = string("linear_58_cast_fp16")]; + string x_125_mode_0 = const()[name = string("x_125_mode_0"), val = string("EXACT")]; + tensor x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = string("x_125_cast_fp16")]; + tensor var_1063_to_fp16 = const()[name = string("op_1063_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(395084480)))]; + tensor var_1064_to_fp16 = const()[name = string("op_1064_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(408191744)))]; + tensor linear_59_cast_fp16 = linear(bias = var_1064_to_fp16, weight = var_1063_to_fp16, x = x_125_cast_fp16)[name = string("linear_59_cast_fp16")]; + tensor x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = string("x_127_cast_fp16")]; + tensor var_1089_axes_0 = const()[name = string("op_1089_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = string("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(408194368)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = string("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(408196992)))]; + fp16 var_1079_to_fp16 = const()[name = string("op_1079_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1089_cast_fp16 = layer_norm(axes = var_1089_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1079_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = string("op_1089_cast_fp16")]; + tensor var_1100_to_fp16 = const()[name = string("op_1100_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(408199616)))]; + tensor var_1101_to_fp16 = const()[name = string("op_1101_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(411476480)))]; + tensor linear_60_cast_fp16 = linear(bias = var_1101_to_fp16, weight = var_1100_to_fp16, x = var_1089_cast_fp16)[name = string("linear_60_cast_fp16")]; + tensor var_1104_to_fp16 = const()[name = string("op_1104_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(411479104)))]; + tensor linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1104_to_fp16, x = var_1089_cast_fp16)[name = string("linear_61_cast_fp16")]; + tensor var_1108_to_fp16 = const()[name = string("op_1108_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(414755968)))]; + tensor var_1109_to_fp16 = const()[name = string("op_1109_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(418032832)))]; + tensor linear_62_cast_fp16 = linear(bias = var_1109_to_fp16, weight = var_1108_to_fp16, x = var_1089_cast_fp16)[name = string("linear_62_cast_fp16")]; + tensor var_1113 = const()[name = string("op_1113"), val = tensor([1, 1500, 20, -1])]; + tensor var_1114_cast_fp16 = reshape(shape = var_1113, x = linear_60_cast_fp16)[name = string("op_1114_cast_fp16")]; + tensor var_1119 = const()[name = string("op_1119"), val = tensor([1, 1500, 20, -1])]; + tensor var_1120_cast_fp16 = reshape(shape = var_1119, x = linear_61_cast_fp16)[name = string("op_1120_cast_fp16")]; + tensor var_1125 = const()[name = string("op_1125"), val = tensor([1, 1500, 20, -1])]; + tensor var_1126_cast_fp16 = reshape(shape = var_1125, x = linear_62_cast_fp16)[name = string("op_1126_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_1126_cast_fp16)[name = string("transpose_165")]; + tensor transpose_127 = transpose(perm = transpose_127_perm_0, x = var_1120_cast_fp16)[name = string("transpose_166")]; + tensor transpose_126 = transpose(perm = transpose_126_perm_0, x = var_1114_cast_fp16)[name = string("transpose_167")]; + tensor a_21_cast_fp16 = scaled_dot_product_attention(key = transpose_127, query = transpose_126, value = transpose_128)[name = string("a_21_cast_fp16")]; + tensor var_1130 = const()[name = string("op_1130"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = string("concat_10"), val = tensor([1, 1500, 1280])]; + tensor var_1131_cast_fp16 = transpose(perm = var_1130, x = a_21_cast_fp16)[name = string("transpose_164")]; + tensor x_131_cast_fp16 = reshape(shape = concat_10, x = var_1131_cast_fp16)[name = string("x_131_cast_fp16")]; + tensor var_1135_to_fp16 = const()[name = string("op_1135_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(418035456)))]; + tensor var_1136_to_fp16 = const()[name = string("op_1136_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(421312320)))]; + tensor linear_63_cast_fp16 = linear(bias = var_1136_to_fp16, weight = var_1135_to_fp16, x = x_131_cast_fp16)[name = string("linear_63_cast_fp16")]; + tensor x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = string("x_133_cast_fp16")]; + tensor var_1143_axes_0 = const()[name = string("op_1143_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = string("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(421314944)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = string("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(421317568)))]; + tensor var_1143_cast_fp16 = layer_norm(axes = var_1143_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1079_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = string("op_1143_cast_fp16")]; + tensor var_1152_to_fp16 = const()[name = string("op_1152_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(421320192)))]; + tensor var_1153_to_fp16 = const()[name = string("op_1153_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(434427456)))]; + tensor linear_64_cast_fp16 = linear(bias = var_1153_to_fp16, weight = var_1152_to_fp16, x = var_1143_cast_fp16)[name = string("linear_64_cast_fp16")]; + string x_137_mode_0 = const()[name = string("x_137_mode_0"), val = string("EXACT")]; + tensor x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = string("x_137_cast_fp16")]; + tensor var_1158_to_fp16 = const()[name = string("op_1158_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(434437760)))]; + tensor var_1159_to_fp16 = const()[name = string("op_1159_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(447545024)))]; + tensor linear_65_cast_fp16 = linear(bias = var_1159_to_fp16, weight = var_1158_to_fp16, x = x_137_cast_fp16)[name = string("linear_65_cast_fp16")]; + tensor x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = string("x_139_cast_fp16")]; + tensor var_1184_axes_0 = const()[name = string("op_1184_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = string("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(447547648)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = string("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(447550272)))]; + fp16 var_1174_to_fp16 = const()[name = string("op_1174_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1184_cast_fp16 = layer_norm(axes = var_1184_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1174_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = string("op_1184_cast_fp16")]; + tensor var_1195_to_fp16 = const()[name = string("op_1195_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(447552896)))]; + tensor var_1196_to_fp16 = const()[name = string("op_1196_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(450829760)))]; + tensor linear_66_cast_fp16 = linear(bias = var_1196_to_fp16, weight = var_1195_to_fp16, x = var_1184_cast_fp16)[name = string("linear_66_cast_fp16")]; + tensor var_1199_to_fp16 = const()[name = string("op_1199_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(450832384)))]; + tensor linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1199_to_fp16, x = var_1184_cast_fp16)[name = string("linear_67_cast_fp16")]; + tensor var_1203_to_fp16 = const()[name = string("op_1203_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(454109248)))]; + tensor var_1204_to_fp16 = const()[name = string("op_1204_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(457386112)))]; + tensor linear_68_cast_fp16 = linear(bias = var_1204_to_fp16, weight = var_1203_to_fp16, x = var_1184_cast_fp16)[name = string("linear_68_cast_fp16")]; + tensor var_1208 = const()[name = string("op_1208"), val = tensor([1, 1500, 20, -1])]; + tensor var_1209_cast_fp16 = reshape(shape = var_1208, x = linear_66_cast_fp16)[name = string("op_1209_cast_fp16")]; + tensor var_1214 = const()[name = string("op_1214"), val = tensor([1, 1500, 20, -1])]; + tensor var_1215_cast_fp16 = reshape(shape = var_1214, x = linear_67_cast_fp16)[name = string("op_1215_cast_fp16")]; + tensor var_1220 = const()[name = string("op_1220"), val = tensor([1, 1500, 20, -1])]; + tensor var_1221_cast_fp16 = reshape(shape = var_1220, x = linear_68_cast_fp16)[name = string("op_1221_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_1221_cast_fp16)[name = string("transpose_161")]; + tensor transpose_130 = transpose(perm = transpose_130_perm_0, x = var_1215_cast_fp16)[name = string("transpose_162")]; + tensor transpose_129 = transpose(perm = transpose_129_perm_0, x = var_1209_cast_fp16)[name = string("transpose_163")]; + tensor a_23_cast_fp16 = scaled_dot_product_attention(key = transpose_130, query = transpose_129, value = transpose_131)[name = string("a_23_cast_fp16")]; + tensor var_1225 = const()[name = string("op_1225"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = string("concat_11"), val = tensor([1, 1500, 1280])]; + tensor var_1226_cast_fp16 = transpose(perm = var_1225, x = a_23_cast_fp16)[name = string("transpose_160")]; + tensor x_143_cast_fp16 = reshape(shape = concat_11, x = var_1226_cast_fp16)[name = string("x_143_cast_fp16")]; + tensor var_1230_to_fp16 = const()[name = string("op_1230_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(457388736)))]; + tensor var_1231_to_fp16 = const()[name = string("op_1231_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(460665600)))]; + tensor linear_69_cast_fp16 = linear(bias = var_1231_to_fp16, weight = var_1230_to_fp16, x = x_143_cast_fp16)[name = string("linear_69_cast_fp16")]; + tensor x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = string("x_145_cast_fp16")]; + tensor var_1238_axes_0 = const()[name = string("op_1238_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = string("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(460668224)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = string("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(460670848)))]; + tensor var_1238_cast_fp16 = layer_norm(axes = var_1238_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1174_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = string("op_1238_cast_fp16")]; + tensor var_1247_to_fp16 = const()[name = string("op_1247_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(460673472)))]; + tensor var_1248_to_fp16 = const()[name = string("op_1248_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(473780736)))]; + tensor linear_70_cast_fp16 = linear(bias = var_1248_to_fp16, weight = var_1247_to_fp16, x = var_1238_cast_fp16)[name = string("linear_70_cast_fp16")]; + string x_149_mode_0 = const()[name = string("x_149_mode_0"), val = string("EXACT")]; + tensor x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = string("x_149_cast_fp16")]; + tensor var_1253_to_fp16 = const()[name = string("op_1253_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(473791040)))]; + tensor var_1254_to_fp16 = const()[name = string("op_1254_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(486898304)))]; + tensor linear_71_cast_fp16 = linear(bias = var_1254_to_fp16, weight = var_1253_to_fp16, x = x_149_cast_fp16)[name = string("linear_71_cast_fp16")]; + tensor x_151_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = string("x_151_cast_fp16")]; + tensor var_1279_axes_0 = const()[name = string("op_1279_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = string("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(486900928)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = string("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(486903552)))]; + fp16 var_1269_to_fp16 = const()[name = string("op_1269_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1279_cast_fp16 = layer_norm(axes = var_1279_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1269_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = string("op_1279_cast_fp16")]; + tensor var_1290_to_fp16 = const()[name = string("op_1290_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(486906176)))]; + tensor var_1291_to_fp16 = const()[name = string("op_1291_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(490183040)))]; + tensor linear_72_cast_fp16 = linear(bias = var_1291_to_fp16, weight = var_1290_to_fp16, x = var_1279_cast_fp16)[name = string("linear_72_cast_fp16")]; + tensor var_1294_to_fp16 = const()[name = string("op_1294_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(490185664)))]; + tensor linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1294_to_fp16, x = var_1279_cast_fp16)[name = string("linear_73_cast_fp16")]; + tensor var_1298_to_fp16 = const()[name = string("op_1298_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(493462528)))]; + tensor var_1299_to_fp16 = const()[name = string("op_1299_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(496739392)))]; + tensor linear_74_cast_fp16 = linear(bias = var_1299_to_fp16, weight = var_1298_to_fp16, x = var_1279_cast_fp16)[name = string("linear_74_cast_fp16")]; + tensor var_1303 = const()[name = string("op_1303"), val = tensor([1, 1500, 20, -1])]; + tensor var_1304_cast_fp16 = reshape(shape = var_1303, x = linear_72_cast_fp16)[name = string("op_1304_cast_fp16")]; + tensor var_1309 = const()[name = string("op_1309"), val = tensor([1, 1500, 20, -1])]; + tensor var_1310_cast_fp16 = reshape(shape = var_1309, x = linear_73_cast_fp16)[name = string("op_1310_cast_fp16")]; + tensor var_1315 = const()[name = string("op_1315"), val = tensor([1, 1500, 20, -1])]; + tensor var_1316_cast_fp16 = reshape(shape = var_1315, x = linear_74_cast_fp16)[name = string("op_1316_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_1316_cast_fp16)[name = string("transpose_157")]; + tensor transpose_133 = transpose(perm = transpose_133_perm_0, x = var_1310_cast_fp16)[name = string("transpose_158")]; + tensor transpose_132 = transpose(perm = transpose_132_perm_0, x = var_1304_cast_fp16)[name = string("transpose_159")]; + tensor a_25_cast_fp16 = scaled_dot_product_attention(key = transpose_133, query = transpose_132, value = transpose_134)[name = string("a_25_cast_fp16")]; + tensor var_1320 = const()[name = string("op_1320"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = string("concat_12"), val = tensor([1, 1500, 1280])]; + tensor var_1321_cast_fp16 = transpose(perm = var_1320, x = a_25_cast_fp16)[name = string("transpose_156")]; + tensor x_155_cast_fp16 = reshape(shape = concat_12, x = var_1321_cast_fp16)[name = string("x_155_cast_fp16")]; + tensor var_1325_to_fp16 = const()[name = string("op_1325_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(496742016)))]; + tensor var_1326_to_fp16 = const()[name = string("op_1326_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(500018880)))]; + tensor linear_75_cast_fp16 = linear(bias = var_1326_to_fp16, weight = var_1325_to_fp16, x = x_155_cast_fp16)[name = string("linear_75_cast_fp16")]; + tensor x_157_cast_fp16 = add(x = x_151_cast_fp16, y = linear_75_cast_fp16)[name = string("x_157_cast_fp16")]; + tensor var_1333_axes_0 = const()[name = string("op_1333_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = string("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(500021504)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = string("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(500024128)))]; + tensor var_1333_cast_fp16 = layer_norm(axes = var_1333_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1269_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = string("op_1333_cast_fp16")]; + tensor var_1342_to_fp16 = const()[name = string("op_1342_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(500026752)))]; + tensor var_1343_to_fp16 = const()[name = string("op_1343_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(513134016)))]; + tensor linear_76_cast_fp16 = linear(bias = var_1343_to_fp16, weight = var_1342_to_fp16, x = var_1333_cast_fp16)[name = string("linear_76_cast_fp16")]; + string x_161_mode_0 = const()[name = string("x_161_mode_0"), val = string("EXACT")]; + tensor x_161_cast_fp16 = gelu(mode = x_161_mode_0, x = linear_76_cast_fp16)[name = string("x_161_cast_fp16")]; + tensor var_1348_to_fp16 = const()[name = string("op_1348_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(513144320)))]; + tensor var_1349_to_fp16 = const()[name = string("op_1349_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(526251584)))]; + tensor linear_77_cast_fp16 = linear(bias = var_1349_to_fp16, weight = var_1348_to_fp16, x = x_161_cast_fp16)[name = string("linear_77_cast_fp16")]; + tensor x_163_cast_fp16 = add(x = x_157_cast_fp16, y = linear_77_cast_fp16)[name = string("x_163_cast_fp16")]; + tensor var_1374_axes_0 = const()[name = string("op_1374_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = string("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(526254208)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = string("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(526256832)))]; + fp16 var_1364_to_fp16 = const()[name = string("op_1364_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1374_cast_fp16 = layer_norm(axes = var_1374_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1364_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = string("op_1374_cast_fp16")]; + tensor var_1385_to_fp16 = const()[name = string("op_1385_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(526259456)))]; + tensor var_1386_to_fp16 = const()[name = string("op_1386_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(529536320)))]; + tensor linear_78_cast_fp16 = linear(bias = var_1386_to_fp16, weight = var_1385_to_fp16, x = var_1374_cast_fp16)[name = string("linear_78_cast_fp16")]; + tensor var_1389_to_fp16 = const()[name = string("op_1389_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(529538944)))]; + tensor linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1389_to_fp16, x = var_1374_cast_fp16)[name = string("linear_79_cast_fp16")]; + tensor var_1393_to_fp16 = const()[name = string("op_1393_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(532815808)))]; + tensor var_1394_to_fp16 = const()[name = string("op_1394_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(536092672)))]; + tensor linear_80_cast_fp16 = linear(bias = var_1394_to_fp16, weight = var_1393_to_fp16, x = var_1374_cast_fp16)[name = string("linear_80_cast_fp16")]; + tensor var_1398 = const()[name = string("op_1398"), val = tensor([1, 1500, 20, -1])]; + tensor var_1399_cast_fp16 = reshape(shape = var_1398, x = linear_78_cast_fp16)[name = string("op_1399_cast_fp16")]; + tensor var_1404 = const()[name = string("op_1404"), val = tensor([1, 1500, 20, -1])]; + tensor var_1405_cast_fp16 = reshape(shape = var_1404, x = linear_79_cast_fp16)[name = string("op_1405_cast_fp16")]; + tensor var_1410 = const()[name = string("op_1410"), val = tensor([1, 1500, 20, -1])]; + tensor var_1411_cast_fp16 = reshape(shape = var_1410, x = linear_80_cast_fp16)[name = string("op_1411_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_1411_cast_fp16)[name = string("transpose_153")]; + tensor transpose_136 = transpose(perm = transpose_136_perm_0, x = var_1405_cast_fp16)[name = string("transpose_154")]; + tensor transpose_135 = transpose(perm = transpose_135_perm_0, x = var_1399_cast_fp16)[name = string("transpose_155")]; + tensor a_27_cast_fp16 = scaled_dot_product_attention(key = transpose_136, query = transpose_135, value = transpose_137)[name = string("a_27_cast_fp16")]; + tensor var_1415 = const()[name = string("op_1415"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = string("concat_13"), val = tensor([1, 1500, 1280])]; + tensor var_1416_cast_fp16 = transpose(perm = var_1415, x = a_27_cast_fp16)[name = string("transpose_152")]; + tensor x_167_cast_fp16 = reshape(shape = concat_13, x = var_1416_cast_fp16)[name = string("x_167_cast_fp16")]; + tensor var_1420_to_fp16 = const()[name = string("op_1420_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(536095296)))]; + tensor var_1421_to_fp16 = const()[name = string("op_1421_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(539372160)))]; + tensor linear_81_cast_fp16 = linear(bias = var_1421_to_fp16, weight = var_1420_to_fp16, x = x_167_cast_fp16)[name = string("linear_81_cast_fp16")]; + tensor x_169_cast_fp16 = add(x = x_163_cast_fp16, y = linear_81_cast_fp16)[name = string("x_169_cast_fp16")]; + tensor var_1428_axes_0 = const()[name = string("op_1428_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = string("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(539374784)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = string("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(539377408)))]; + tensor var_1428_cast_fp16 = layer_norm(axes = var_1428_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1364_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = string("op_1428_cast_fp16")]; + tensor var_1437_to_fp16 = const()[name = string("op_1437_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(539380032)))]; + tensor var_1438_to_fp16 = const()[name = string("op_1438_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(552487296)))]; + tensor linear_82_cast_fp16 = linear(bias = var_1438_to_fp16, weight = var_1437_to_fp16, x = var_1428_cast_fp16)[name = string("linear_82_cast_fp16")]; + string x_173_mode_0 = const()[name = string("x_173_mode_0"), val = string("EXACT")]; + tensor x_173_cast_fp16 = gelu(mode = x_173_mode_0, x = linear_82_cast_fp16)[name = string("x_173_cast_fp16")]; + tensor var_1443_to_fp16 = const()[name = string("op_1443_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(552497600)))]; + tensor var_1444_to_fp16 = const()[name = string("op_1444_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(565604864)))]; + tensor linear_83_cast_fp16 = linear(bias = var_1444_to_fp16, weight = var_1443_to_fp16, x = x_173_cast_fp16)[name = string("linear_83_cast_fp16")]; + tensor x_175_cast_fp16 = add(x = x_169_cast_fp16, y = linear_83_cast_fp16)[name = string("x_175_cast_fp16")]; + tensor var_1469_axes_0 = const()[name = string("op_1469_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = string("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(565607488)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = string("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(565610112)))]; + fp16 var_1459_to_fp16 = const()[name = string("op_1459_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1469_cast_fp16 = layer_norm(axes = var_1469_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1459_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = string("op_1469_cast_fp16")]; + tensor var_1480_to_fp16 = const()[name = string("op_1480_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(565612736)))]; + tensor var_1481_to_fp16 = const()[name = string("op_1481_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(568889600)))]; + tensor linear_84_cast_fp16 = linear(bias = var_1481_to_fp16, weight = var_1480_to_fp16, x = var_1469_cast_fp16)[name = string("linear_84_cast_fp16")]; + tensor var_1484_to_fp16 = const()[name = string("op_1484_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(568892224)))]; + tensor linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1484_to_fp16, x = var_1469_cast_fp16)[name = string("linear_85_cast_fp16")]; + tensor var_1488_to_fp16 = const()[name = string("op_1488_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(572169088)))]; + tensor var_1489_to_fp16 = const()[name = string("op_1489_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(575445952)))]; + tensor linear_86_cast_fp16 = linear(bias = var_1489_to_fp16, weight = var_1488_to_fp16, x = var_1469_cast_fp16)[name = string("linear_86_cast_fp16")]; + tensor var_1493 = const()[name = string("op_1493"), val = tensor([1, 1500, 20, -1])]; + tensor var_1494_cast_fp16 = reshape(shape = var_1493, x = linear_84_cast_fp16)[name = string("op_1494_cast_fp16")]; + tensor var_1499 = const()[name = string("op_1499"), val = tensor([1, 1500, 20, -1])]; + tensor var_1500_cast_fp16 = reshape(shape = var_1499, x = linear_85_cast_fp16)[name = string("op_1500_cast_fp16")]; + tensor var_1505 = const()[name = string("op_1505"), val = tensor([1, 1500, 20, -1])]; + tensor var_1506_cast_fp16 = reshape(shape = var_1505, x = linear_86_cast_fp16)[name = string("op_1506_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_1506_cast_fp16)[name = string("transpose_149")]; + tensor transpose_139 = transpose(perm = transpose_139_perm_0, x = var_1500_cast_fp16)[name = string("transpose_150")]; + tensor transpose_138 = transpose(perm = transpose_138_perm_0, x = var_1494_cast_fp16)[name = string("transpose_151")]; + tensor a_29_cast_fp16 = scaled_dot_product_attention(key = transpose_139, query = transpose_138, value = transpose_140)[name = string("a_29_cast_fp16")]; + tensor var_1510 = const()[name = string("op_1510"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = string("concat_14"), val = tensor([1, 1500, 1280])]; + tensor var_1511_cast_fp16 = transpose(perm = var_1510, x = a_29_cast_fp16)[name = string("transpose_148")]; + tensor x_179_cast_fp16 = reshape(shape = concat_14, x = var_1511_cast_fp16)[name = string("x_179_cast_fp16")]; + tensor var_1515_to_fp16 = const()[name = string("op_1515_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(575448576)))]; + tensor var_1516_to_fp16 = const()[name = string("op_1516_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(578725440)))]; + tensor linear_87_cast_fp16 = linear(bias = var_1516_to_fp16, weight = var_1515_to_fp16, x = x_179_cast_fp16)[name = string("linear_87_cast_fp16")]; + tensor x_181_cast_fp16 = add(x = x_175_cast_fp16, y = linear_87_cast_fp16)[name = string("x_181_cast_fp16")]; + tensor var_1523_axes_0 = const()[name = string("op_1523_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = string("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(578728064)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = string("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(578730688)))]; + tensor var_1523_cast_fp16 = layer_norm(axes = var_1523_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1459_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = string("op_1523_cast_fp16")]; + tensor var_1532_to_fp16 = const()[name = string("op_1532_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(578733312)))]; + tensor var_1533_to_fp16 = const()[name = string("op_1533_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(591840576)))]; + tensor linear_88_cast_fp16 = linear(bias = var_1533_to_fp16, weight = var_1532_to_fp16, x = var_1523_cast_fp16)[name = string("linear_88_cast_fp16")]; + string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("EXACT")]; + tensor x_185_cast_fp16 = gelu(mode = x_185_mode_0, x = linear_88_cast_fp16)[name = string("x_185_cast_fp16")]; + tensor var_1538_to_fp16 = const()[name = string("op_1538_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(591850880)))]; + tensor var_1539_to_fp16 = const()[name = string("op_1539_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(604958144)))]; + tensor linear_89_cast_fp16 = linear(bias = var_1539_to_fp16, weight = var_1538_to_fp16, x = x_185_cast_fp16)[name = string("linear_89_cast_fp16")]; + tensor x_187_cast_fp16 = add(x = x_181_cast_fp16, y = linear_89_cast_fp16)[name = string("x_187_cast_fp16")]; + tensor var_1564_axes_0 = const()[name = string("op_1564_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = string("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(604960768)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = string("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(604963392)))]; + fp16 var_1554_to_fp16 = const()[name = string("op_1554_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1564_cast_fp16 = layer_norm(axes = var_1564_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1554_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = string("op_1564_cast_fp16")]; + tensor var_1575_to_fp16 = const()[name = string("op_1575_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(604966016)))]; + tensor var_1576_to_fp16 = const()[name = string("op_1576_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(608242880)))]; + tensor linear_90_cast_fp16 = linear(bias = var_1576_to_fp16, weight = var_1575_to_fp16, x = var_1564_cast_fp16)[name = string("linear_90_cast_fp16")]; + tensor var_1579_to_fp16 = const()[name = string("op_1579_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(608245504)))]; + tensor linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1579_to_fp16, x = var_1564_cast_fp16)[name = string("linear_91_cast_fp16")]; + tensor var_1583_to_fp16 = const()[name = string("op_1583_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(611522368)))]; + tensor var_1584_to_fp16 = const()[name = string("op_1584_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(614799232)))]; + tensor linear_92_cast_fp16 = linear(bias = var_1584_to_fp16, weight = var_1583_to_fp16, x = var_1564_cast_fp16)[name = string("linear_92_cast_fp16")]; + tensor var_1588 = const()[name = string("op_1588"), val = tensor([1, 1500, 20, -1])]; + tensor var_1589_cast_fp16 = reshape(shape = var_1588, x = linear_90_cast_fp16)[name = string("op_1589_cast_fp16")]; + tensor var_1594 = const()[name = string("op_1594"), val = tensor([1, 1500, 20, -1])]; + tensor var_1595_cast_fp16 = reshape(shape = var_1594, x = linear_91_cast_fp16)[name = string("op_1595_cast_fp16")]; + tensor var_1600 = const()[name = string("op_1600"), val = tensor([1, 1500, 20, -1])]; + tensor var_1601_cast_fp16 = reshape(shape = var_1600, x = linear_92_cast_fp16)[name = string("op_1601_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_1601_cast_fp16)[name = string("transpose_145")]; + tensor transpose_142 = transpose(perm = transpose_142_perm_0, x = var_1595_cast_fp16)[name = string("transpose_146")]; + tensor transpose_141 = transpose(perm = transpose_141_perm_0, x = var_1589_cast_fp16)[name = string("transpose_147")]; + tensor a_31_cast_fp16 = scaled_dot_product_attention(key = transpose_142, query = transpose_141, value = transpose_143)[name = string("a_31_cast_fp16")]; + tensor var_1605 = const()[name = string("op_1605"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = string("concat_15"), val = tensor([1, 1500, 1280])]; + tensor var_1606_cast_fp16 = transpose(perm = var_1605, x = a_31_cast_fp16)[name = string("transpose_144")]; + tensor x_191_cast_fp16 = reshape(shape = concat_15, x = var_1606_cast_fp16)[name = string("x_191_cast_fp16")]; + tensor var_1610_to_fp16 = const()[name = string("op_1610_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(614801856)))]; + tensor var_1611_to_fp16 = const()[name = string("op_1611_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(618078720)))]; + tensor linear_93_cast_fp16 = linear(bias = var_1611_to_fp16, weight = var_1610_to_fp16, x = x_191_cast_fp16)[name = string("linear_93_cast_fp16")]; + tensor x_193_cast_fp16_1 = add(x = x_187_cast_fp16, y = linear_93_cast_fp16)[name = string("x_193_cast_fp16")]; + tensor var_1618_axes_0 = const()[name = string("op_1618_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = string("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(618081344)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = string("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(618083968)))]; + tensor var_1618_cast_fp16 = layer_norm(axes = var_1618_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1554_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16_1)[name = string("op_1618_cast_fp16")]; + tensor var_1627_to_fp16 = const()[name = string("op_1627_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(618086592)))]; + tensor var_1628_to_fp16 = const()[name = string("op_1628_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(631193856)))]; + tensor linear_94_cast_fp16 = linear(bias = var_1628_to_fp16, weight = var_1627_to_fp16, x = var_1618_cast_fp16)[name = string("linear_94_cast_fp16")]; + string x_197_mode_0 = const()[name = string("x_197_mode_0"), val = string("EXACT")]; + tensor x_197_cast_fp16 = gelu(mode = x_197_mode_0, x = linear_94_cast_fp16)[name = string("x_197_cast_fp16")]; + tensor var_1633_to_fp16 = const()[name = string("op_1633_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(631204160)))]; + tensor var_1634_to_fp16 = const()[name = string("op_1634_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(644311424)))]; + tensor linear_95_cast_fp16_1 = linear(bias = var_1634_to_fp16, weight = var_1633_to_fp16, x = x_197_cast_fp16)[name = string("linear_95_cast_fp16")]; + string linear_95_cast_fp16_dtype_0 = const()[name = string("linear_95_cast_fp16_dtype_0"), val = string("fp32")]; + string x_193_cast_fp16_dtype_0 = const()[name = string("x_193_cast_fp16_dtype_0"), val = string("fp32")]; + tensor x_193_cast_fp16 = cast(dtype = x_193_cast_fp16_dtype_0, x = x_193_cast_fp16_1)[name = string("cast_2")]; + tensor linear_95_cast_fp16 = cast(dtype = linear_95_cast_fp16_dtype_0, x = linear_95_cast_fp16_1)[name = string("cast_3")]; + } -> (linear_95_cast_fp16, x_193_cast_fp16); +} \ No newline at end of file