diff --git "a/ggml-medium-encoder.mlmodelc/model.mil" "b/ggml-medium-encoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/ggml-medium-encoder.mlmodelc/model.mil" @@ -0,0 +1,1428 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}, {"coremltools-component-torch", "2.2.2"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.2"}})] +{ + func main(tensor logmel_data) { + tensor var_56 = const()[name = tensor("op_56"), val = tensor(1)]; + tensor var_64 = const()[name = tensor("op_64"), val = tensor([1])]; + tensor var_66 = const()[name = tensor("op_66"), val = tensor([1])]; + tensor var_68_pad_type_0 = const()[name = tensor("op_68_pad_type_0"), val = tensor("custom")]; + tensor var_68_pad_0 = const()[name = tensor("op_68_pad_0"), val = tensor([1, 1])]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491648)))]; + tensor var_68_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_66, groups = var_56, pad = var_68_pad_0, pad_type = var_68_pad_type_0, strides = var_64, weight = weight_3_to_fp16, x = logmel_data)[name = tensor("op_68_cast_fp16")]; + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_68_cast_fp16)[name = tensor("input_1_cast_fp16")]; + tensor var_73 = const()[name = tensor("op_73"), val = tensor(1)]; + tensor var_82 = const()[name = tensor("op_82"), val = tensor([2])]; + tensor var_84 = const()[name = tensor("op_84"), val = tensor([1])]; + tensor var_86_pad_type_0 = const()[name = tensor("op_86_pad_type_0"), val = tensor("custom")]; + tensor var_86_pad_0 = const()[name = tensor("op_86_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493760)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6785280)))]; + tensor var_86_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_84, groups = var_73, pad = var_86_pad_0, pad_type = var_86_pad_type_0, strides = var_82, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = tensor("op_86_cast_fp16")]; + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_86_cast_fp16)[name = tensor("x_3_cast_fp16")]; + tensor var_92 = const()[name = tensor("op_92"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6787392)))]; + tensor transpose_240 = transpose(perm = var_92, x = x_3_cast_fp16)[name = tensor("transpose_240")]; + tensor var_95_cast_fp16 = add(x = transpose_240, y = positional_embedding_to_fp16)[name = tensor("op_95_cast_fp16")]; + tensor var_108 = const()[name = tensor("op_108"), val = tensor(-1)]; + tensor var_124_axes_0 = const()[name = tensor("op_124_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9859456)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9861568)))]; + tensor var_114_to_fp16 = const()[name = tensor("op_114_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_124_cast_fp16 = layer_norm(axes = var_124_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_114_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_95_cast_fp16)[name = tensor("op_124_cast_fp16")]; + tensor var_135_to_fp16 = const()[name = tensor("op_135_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9863680)))]; + tensor var_136_to_fp16 = const()[name = tensor("op_136_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11960896)))]; + tensor linear_0_cast_fp16 = linear(bias = var_136_to_fp16, weight = var_135_to_fp16, x = var_124_cast_fp16)[name = tensor("linear_0_cast_fp16")]; + tensor var_139_to_fp16 = const()[name = tensor("op_139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11963008)))]; + tensor linear_1_bias_0_to_fp16 = const()[name = tensor("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14060224)))]; + tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_139_to_fp16, x = var_124_cast_fp16)[name = tensor("linear_1_cast_fp16")]; + tensor var_143_to_fp16 = const()[name = tensor("op_143_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14062336)))]; + tensor var_144_to_fp16 = const()[name = tensor("op_144_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16159552)))]; + tensor linear_2_cast_fp16 = linear(bias = var_144_to_fp16, weight = var_143_to_fp16, x = var_124_cast_fp16)[name = tensor("linear_2_cast_fp16")]; + tensor var_152 = const()[name = tensor("op_152"), val = tensor([1, 1500, 16, -1])]; + tensor var_153_cast_fp16 = reshape(shape = var_152, x = linear_0_cast_fp16)[name = tensor("op_153_cast_fp16")]; + tensor const_168_to_fp16 = const()[name = tensor("const_168_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast_fp16 = mul(x = var_153_cast_fp16, y = const_168_to_fp16)[name = tensor("q_3_cast_fp16")]; + tensor var_159 = const()[name = tensor("op_159"), val = tensor([1, 1500, 16, -1])]; + tensor var_160_cast_fp16 = reshape(shape = var_159, x = linear_1_cast_fp16)[name = tensor("op_160_cast_fp16")]; + tensor const_169_to_fp16 = const()[name = tensor("const_169_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast_fp16 = mul(x = var_160_cast_fp16, y = const_169_to_fp16)[name = tensor("k_3_cast_fp16")]; + tensor var_166 = const()[name = tensor("op_166"), val = tensor([1, 1500, 16, -1])]; + tensor var_167_cast_fp16 = reshape(shape = var_166, x = linear_2_cast_fp16)[name = tensor("op_167_cast_fp16")]; + tensor var_168 = const()[name = tensor("op_168"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_96_perm_0 = const()[name = tensor("transpose_96_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_97_perm_0 = const()[name = tensor("transpose_97_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_237 = transpose(perm = transpose_97_perm_0, x = k_3_cast_fp16)[name = tensor("transpose_237")]; + tensor transpose_238 = transpose(perm = transpose_96_perm_0, x = q_3_cast_fp16)[name = tensor("transpose_238")]; + tensor qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_238, y = transpose_237)[name = tensor("qk_1_cast_fp16")]; + tensor var_172_cast_fp16 = softmax(axis = var_108, x = qk_1_cast_fp16)[name = tensor("op_172_cast_fp16")]; + tensor var_174_transpose_x_0 = const()[name = tensor("op_174_transpose_x_0"), val = tensor(false)]; + tensor var_174_transpose_y_0 = const()[name = tensor("op_174_transpose_y_0"), val = tensor(false)]; + tensor transpose_239 = transpose(perm = var_168, x = var_167_cast_fp16)[name = tensor("transpose_239")]; + tensor var_174_cast_fp16 = matmul(transpose_x = var_174_transpose_x_0, transpose_y = var_174_transpose_y_0, x = var_172_cast_fp16, y = transpose_239)[name = tensor("op_174_cast_fp16")]; + tensor var_175 = const()[name = tensor("op_175"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 1024])]; + tensor transpose_236 = transpose(perm = var_175, x = var_174_cast_fp16)[name = tensor("transpose_236")]; + tensor x_11_cast_fp16 = reshape(shape = concat_0, x = transpose_236)[name = tensor("x_11_cast_fp16")]; + tensor var_180_to_fp16 = const()[name = tensor("op_180_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16161664)))]; + tensor var_181_to_fp16 = const()[name = tensor("op_181_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18258880)))]; + tensor linear_3_cast_fp16 = linear(bias = var_181_to_fp16, weight = var_180_to_fp16, x = x_11_cast_fp16)[name = tensor("linear_3_cast_fp16")]; + tensor x_13_cast_fp16 = add(x = var_95_cast_fp16, y = linear_3_cast_fp16)[name = tensor("x_13_cast_fp16")]; + tensor var_188_axes_0 = const()[name = tensor("op_188_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18260992)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18263104)))]; + tensor var_188_cast_fp16 = layer_norm(axes = var_188_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_114_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor("op_188_cast_fp16")]; + tensor var_197_to_fp16 = const()[name = tensor("op_197_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18265216)))]; + tensor var_198_to_fp16 = const()[name = tensor("op_198_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26653888)))]; + tensor linear_4_cast_fp16 = linear(bias = var_198_to_fp16, weight = var_197_to_fp16, x = var_188_cast_fp16)[name = tensor("linear_4_cast_fp16")]; + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = tensor("x_17_cast_fp16")]; + tensor var_203_to_fp16 = const()[name = tensor("op_203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26662144)))]; + tensor var_204_to_fp16 = const()[name = tensor("op_204_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35050816)))]; + tensor linear_5_cast_fp16 = linear(bias = var_204_to_fp16, weight = var_203_to_fp16, x = x_17_cast_fp16)[name = tensor("linear_5_cast_fp16")]; + tensor x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = tensor("x_19_cast_fp16")]; + tensor var_214 = const()[name = tensor("op_214"), val = tensor(-1)]; + tensor var_230_axes_0 = const()[name = tensor("op_230_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35052928)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35055040)))]; + tensor var_220_to_fp16 = const()[name = tensor("op_220_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_230_cast_fp16 = layer_norm(axes = var_230_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_220_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = tensor("op_230_cast_fp16")]; + tensor var_241_to_fp16 = const()[name = tensor("op_241_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35057152)))]; + tensor var_242_to_fp16 = const()[name = tensor("op_242_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37154368)))]; + tensor linear_6_cast_fp16 = linear(bias = var_242_to_fp16, weight = var_241_to_fp16, x = var_230_cast_fp16)[name = tensor("linear_6_cast_fp16")]; + tensor var_245_to_fp16 = const()[name = tensor("op_245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37156480)))]; + tensor linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_245_to_fp16, x = var_230_cast_fp16)[name = tensor("linear_7_cast_fp16")]; + tensor var_249_to_fp16 = const()[name = tensor("op_249_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39253696)))]; + tensor var_250_to_fp16 = const()[name = tensor("op_250_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41350912)))]; + tensor linear_8_cast_fp16 = linear(bias = var_250_to_fp16, weight = var_249_to_fp16, x = var_230_cast_fp16)[name = tensor("linear_8_cast_fp16")]; + tensor var_258 = const()[name = tensor("op_258"), val = tensor([1, 1500, 16, -1])]; + tensor var_259_cast_fp16 = reshape(shape = var_258, x = linear_6_cast_fp16)[name = tensor("op_259_cast_fp16")]; + tensor const_170_to_fp16 = const()[name = tensor("const_170_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast_fp16 = mul(x = var_259_cast_fp16, y = const_170_to_fp16)[name = tensor("q_7_cast_fp16")]; + tensor var_265 = const()[name = tensor("op_265"), val = tensor([1, 1500, 16, -1])]; + tensor var_266_cast_fp16 = reshape(shape = var_265, x = linear_7_cast_fp16)[name = tensor("op_266_cast_fp16")]; + tensor const_171_to_fp16 = const()[name = tensor("const_171_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast_fp16 = mul(x = var_266_cast_fp16, y = const_171_to_fp16)[name = tensor("k_7_cast_fp16")]; + tensor var_272 = const()[name = tensor("op_272"), val = tensor([1, 1500, 16, -1])]; + tensor var_273_cast_fp16 = reshape(shape = var_272, x = linear_8_cast_fp16)[name = tensor("op_273_cast_fp16")]; + tensor var_274 = const()[name = tensor("op_274"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_98_perm_0 = const()[name = tensor("transpose_98_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_99_perm_0 = const()[name = tensor("transpose_99_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_233 = transpose(perm = transpose_99_perm_0, x = k_7_cast_fp16)[name = tensor("transpose_233")]; + tensor transpose_234 = transpose(perm = transpose_98_perm_0, x = q_7_cast_fp16)[name = tensor("transpose_234")]; + tensor qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_234, y = transpose_233)[name = tensor("qk_3_cast_fp16")]; + tensor var_278_cast_fp16 = softmax(axis = var_214, x = qk_3_cast_fp16)[name = tensor("op_278_cast_fp16")]; + tensor var_280_transpose_x_0 = const()[name = tensor("op_280_transpose_x_0"), val = tensor(false)]; + tensor var_280_transpose_y_0 = const()[name = tensor("op_280_transpose_y_0"), val = tensor(false)]; + tensor transpose_235 = transpose(perm = var_274, x = var_273_cast_fp16)[name = tensor("transpose_235")]; + tensor var_280_cast_fp16 = matmul(transpose_x = var_280_transpose_x_0, transpose_y = var_280_transpose_y_0, x = var_278_cast_fp16, y = transpose_235)[name = tensor("op_280_cast_fp16")]; + tensor var_281 = const()[name = tensor("op_281"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 1024])]; + tensor transpose_232 = transpose(perm = var_281, x = var_280_cast_fp16)[name = tensor("transpose_232")]; + tensor x_23_cast_fp16 = reshape(shape = concat_1, x = transpose_232)[name = tensor("x_23_cast_fp16")]; + tensor var_286_to_fp16 = const()[name = tensor("op_286_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41353024)))]; + tensor var_287_to_fp16 = const()[name = tensor("op_287_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43450240)))]; + tensor linear_9_cast_fp16 = linear(bias = var_287_to_fp16, weight = var_286_to_fp16, x = x_23_cast_fp16)[name = tensor("linear_9_cast_fp16")]; + tensor x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor("x_25_cast_fp16")]; + tensor var_294_axes_0 = const()[name = tensor("op_294_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43452352)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43454464)))]; + tensor var_294_cast_fp16 = layer_norm(axes = var_294_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_220_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = tensor("op_294_cast_fp16")]; + tensor var_303_to_fp16 = const()[name = tensor("op_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43456576)))]; + tensor var_304_to_fp16 = const()[name = tensor("op_304_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51845248)))]; + tensor linear_10_cast_fp16 = linear(bias = var_304_to_fp16, weight = var_303_to_fp16, x = var_294_cast_fp16)[name = tensor("linear_10_cast_fp16")]; + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = tensor("x_29_cast_fp16")]; + tensor var_309_to_fp16 = const()[name = tensor("op_309_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51853504)))]; + tensor var_310_to_fp16 = const()[name = tensor("op_310_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60242176)))]; + tensor linear_11_cast_fp16 = linear(bias = var_310_to_fp16, weight = var_309_to_fp16, x = x_29_cast_fp16)[name = tensor("linear_11_cast_fp16")]; + tensor x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = tensor("x_31_cast_fp16")]; + tensor var_320 = const()[name = tensor("op_320"), val = tensor(-1)]; + tensor var_336_axes_0 = const()[name = tensor("op_336_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60244288)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60246400)))]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_336_cast_fp16 = layer_norm(axes = var_336_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_326_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = tensor("op_336_cast_fp16")]; + tensor var_347_to_fp16 = const()[name = tensor("op_347_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60248512)))]; + tensor var_348_to_fp16 = const()[name = tensor("op_348_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62345728)))]; + tensor linear_12_cast_fp16 = linear(bias = var_348_to_fp16, weight = var_347_to_fp16, x = var_336_cast_fp16)[name = tensor("linear_12_cast_fp16")]; + tensor var_351_to_fp16 = const()[name = tensor("op_351_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62347840)))]; + tensor linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_351_to_fp16, x = var_336_cast_fp16)[name = tensor("linear_13_cast_fp16")]; + tensor var_355_to_fp16 = const()[name = tensor("op_355_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64445056)))]; + tensor var_356_to_fp16 = const()[name = tensor("op_356_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66542272)))]; + tensor linear_14_cast_fp16 = linear(bias = var_356_to_fp16, weight = var_355_to_fp16, x = var_336_cast_fp16)[name = tensor("linear_14_cast_fp16")]; + tensor var_364 = const()[name = tensor("op_364"), val = tensor([1, 1500, 16, -1])]; + tensor var_365_cast_fp16 = reshape(shape = var_364, x = linear_12_cast_fp16)[name = tensor("op_365_cast_fp16")]; + tensor const_172_to_fp16 = const()[name = tensor("const_172_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast_fp16 = mul(x = var_365_cast_fp16, y = const_172_to_fp16)[name = tensor("q_11_cast_fp16")]; + tensor var_371 = const()[name = tensor("op_371"), val = tensor([1, 1500, 16, -1])]; + tensor var_372_cast_fp16 = reshape(shape = var_371, x = linear_13_cast_fp16)[name = tensor("op_372_cast_fp16")]; + tensor const_173_to_fp16 = const()[name = tensor("const_173_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast_fp16 = mul(x = var_372_cast_fp16, y = const_173_to_fp16)[name = tensor("k_11_cast_fp16")]; + tensor var_378 = const()[name = tensor("op_378"), val = tensor([1, 1500, 16, -1])]; + tensor var_379_cast_fp16 = reshape(shape = var_378, x = linear_14_cast_fp16)[name = tensor("op_379_cast_fp16")]; + tensor var_380 = const()[name = tensor("op_380"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_100_perm_0 = const()[name = tensor("transpose_100_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_101_perm_0 = const()[name = tensor("transpose_101_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_229 = transpose(perm = transpose_101_perm_0, x = k_11_cast_fp16)[name = tensor("transpose_229")]; + tensor transpose_230 = transpose(perm = transpose_100_perm_0, x = q_11_cast_fp16)[name = tensor("transpose_230")]; + tensor qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_230, y = transpose_229)[name = tensor("qk_5_cast_fp16")]; + tensor var_384_cast_fp16 = softmax(axis = var_320, x = qk_5_cast_fp16)[name = tensor("op_384_cast_fp16")]; + tensor var_386_transpose_x_0 = const()[name = tensor("op_386_transpose_x_0"), val = tensor(false)]; + tensor var_386_transpose_y_0 = const()[name = tensor("op_386_transpose_y_0"), val = tensor(false)]; + tensor transpose_231 = transpose(perm = var_380, x = var_379_cast_fp16)[name = tensor("transpose_231")]; + tensor var_386_cast_fp16 = matmul(transpose_x = var_386_transpose_x_0, transpose_y = var_386_transpose_y_0, x = var_384_cast_fp16, y = transpose_231)[name = tensor("op_386_cast_fp16")]; + tensor var_387 = const()[name = tensor("op_387"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 1024])]; + tensor transpose_228 = transpose(perm = var_387, x = var_386_cast_fp16)[name = tensor("transpose_228")]; + tensor x_35_cast_fp16 = reshape(shape = concat_2, x = transpose_228)[name = tensor("x_35_cast_fp16")]; + tensor var_392_to_fp16 = const()[name = tensor("op_392_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66544384)))]; + tensor var_393_to_fp16 = const()[name = tensor("op_393_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68641600)))]; + tensor linear_15_cast_fp16 = linear(bias = var_393_to_fp16, weight = var_392_to_fp16, x = x_35_cast_fp16)[name = tensor("linear_15_cast_fp16")]; + tensor x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = tensor("x_37_cast_fp16")]; + tensor var_400_axes_0 = const()[name = tensor("op_400_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68643712)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68645824)))]; + tensor var_400_cast_fp16 = layer_norm(axes = var_400_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_326_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = tensor("op_400_cast_fp16")]; + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68647936)))]; + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77036608)))]; + tensor linear_16_cast_fp16 = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = var_400_cast_fp16)[name = tensor("linear_16_cast_fp16")]; + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = tensor("x_41_cast_fp16")]; + tensor var_415_to_fp16 = const()[name = tensor("op_415_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77044864)))]; + tensor var_416_to_fp16 = const()[name = tensor("op_416_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85433536)))]; + tensor linear_17_cast_fp16 = linear(bias = var_416_to_fp16, weight = var_415_to_fp16, x = x_41_cast_fp16)[name = tensor("linear_17_cast_fp16")]; + tensor x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = tensor("x_43_cast_fp16")]; + tensor var_426 = const()[name = tensor("op_426"), val = tensor(-1)]; + tensor var_442_axes_0 = const()[name = tensor("op_442_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85435648)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85437760)))]; + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_442_cast_fp16 = layer_norm(axes = var_442_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_432_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = tensor("op_442_cast_fp16")]; + tensor var_453_to_fp16 = const()[name = tensor("op_453_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85439872)))]; + tensor var_454_to_fp16 = const()[name = tensor("op_454_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87537088)))]; + tensor linear_18_cast_fp16 = linear(bias = var_454_to_fp16, weight = var_453_to_fp16, x = var_442_cast_fp16)[name = tensor("linear_18_cast_fp16")]; + tensor var_457_to_fp16 = const()[name = tensor("op_457_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87539200)))]; + tensor linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_457_to_fp16, x = var_442_cast_fp16)[name = tensor("linear_19_cast_fp16")]; + tensor var_461_to_fp16 = const()[name = tensor("op_461_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89636416)))]; + tensor var_462_to_fp16 = const()[name = tensor("op_462_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91733632)))]; + tensor linear_20_cast_fp16 = linear(bias = var_462_to_fp16, weight = var_461_to_fp16, x = var_442_cast_fp16)[name = tensor("linear_20_cast_fp16")]; + tensor var_470 = const()[name = tensor("op_470"), val = tensor([1, 1500, 16, -1])]; + tensor var_471_cast_fp16 = reshape(shape = var_470, x = linear_18_cast_fp16)[name = tensor("op_471_cast_fp16")]; + tensor const_174_to_fp16 = const()[name = tensor("const_174_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast_fp16 = mul(x = var_471_cast_fp16, y = const_174_to_fp16)[name = tensor("q_15_cast_fp16")]; + tensor var_477 = const()[name = tensor("op_477"), val = tensor([1, 1500, 16, -1])]; + tensor var_478_cast_fp16 = reshape(shape = var_477, x = linear_19_cast_fp16)[name = tensor("op_478_cast_fp16")]; + tensor const_175_to_fp16 = const()[name = tensor("const_175_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast_fp16 = mul(x = var_478_cast_fp16, y = const_175_to_fp16)[name = tensor("k_15_cast_fp16")]; + tensor var_484 = const()[name = tensor("op_484"), val = tensor([1, 1500, 16, -1])]; + tensor var_485_cast_fp16 = reshape(shape = var_484, x = linear_20_cast_fp16)[name = tensor("op_485_cast_fp16")]; + tensor var_486 = const()[name = tensor("op_486"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_102_perm_0 = const()[name = tensor("transpose_102_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_103_perm_0 = const()[name = tensor("transpose_103_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_225 = transpose(perm = transpose_103_perm_0, x = k_15_cast_fp16)[name = tensor("transpose_225")]; + tensor transpose_226 = transpose(perm = transpose_102_perm_0, x = q_15_cast_fp16)[name = tensor("transpose_226")]; + tensor qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_226, y = transpose_225)[name = tensor("qk_7_cast_fp16")]; + tensor var_490_cast_fp16 = softmax(axis = var_426, x = qk_7_cast_fp16)[name = tensor("op_490_cast_fp16")]; + tensor var_492_transpose_x_0 = const()[name = tensor("op_492_transpose_x_0"), val = tensor(false)]; + tensor var_492_transpose_y_0 = const()[name = tensor("op_492_transpose_y_0"), val = tensor(false)]; + tensor transpose_227 = transpose(perm = var_486, x = var_485_cast_fp16)[name = tensor("transpose_227")]; + tensor var_492_cast_fp16 = matmul(transpose_x = var_492_transpose_x_0, transpose_y = var_492_transpose_y_0, x = var_490_cast_fp16, y = transpose_227)[name = tensor("op_492_cast_fp16")]; + tensor var_493 = const()[name = tensor("op_493"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 1024])]; + tensor transpose_224 = transpose(perm = var_493, x = var_492_cast_fp16)[name = tensor("transpose_224")]; + tensor x_47_cast_fp16 = reshape(shape = concat_3, x = transpose_224)[name = tensor("x_47_cast_fp16")]; + tensor var_498_to_fp16 = const()[name = tensor("op_498_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91735744)))]; + tensor var_499_to_fp16 = const()[name = tensor("op_499_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93832960)))]; + tensor linear_21_cast_fp16 = linear(bias = var_499_to_fp16, weight = var_498_to_fp16, x = x_47_cast_fp16)[name = tensor("linear_21_cast_fp16")]; + tensor x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = tensor("x_49_cast_fp16")]; + tensor var_506_axes_0 = const()[name = tensor("op_506_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93835072)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93837184)))]; + tensor var_506_cast_fp16 = layer_norm(axes = var_506_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_432_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = tensor("op_506_cast_fp16")]; + tensor var_515_to_fp16 = const()[name = tensor("op_515_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93839296)))]; + tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102227968)))]; + tensor linear_22_cast_fp16 = linear(bias = var_516_to_fp16, weight = var_515_to_fp16, x = var_506_cast_fp16)[name = tensor("linear_22_cast_fp16")]; + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = tensor("x_53_cast_fp16")]; + tensor var_521_to_fp16 = const()[name = tensor("op_521_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102236224)))]; + tensor var_522_to_fp16 = const()[name = tensor("op_522_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110624896)))]; + tensor linear_23_cast_fp16 = linear(bias = var_522_to_fp16, weight = var_521_to_fp16, x = x_53_cast_fp16)[name = tensor("linear_23_cast_fp16")]; + tensor x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = tensor("x_55_cast_fp16")]; + tensor var_532 = const()[name = tensor("op_532"), val = tensor(-1)]; + tensor var_548_axes_0 = const()[name = tensor("op_548_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110627008)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110629120)))]; + tensor var_538_to_fp16 = const()[name = tensor("op_538_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_548_cast_fp16 = layer_norm(axes = var_548_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_538_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = tensor("op_548_cast_fp16")]; + tensor var_559_to_fp16 = const()[name = tensor("op_559_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110631232)))]; + tensor var_560_to_fp16 = const()[name = tensor("op_560_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112728448)))]; + tensor linear_24_cast_fp16 = linear(bias = var_560_to_fp16, weight = var_559_to_fp16, x = var_548_cast_fp16)[name = tensor("linear_24_cast_fp16")]; + tensor var_563_to_fp16 = const()[name = tensor("op_563_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112730560)))]; + tensor linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_563_to_fp16, x = var_548_cast_fp16)[name = tensor("linear_25_cast_fp16")]; + tensor var_567_to_fp16 = const()[name = tensor("op_567_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114827776)))]; + tensor var_568_to_fp16 = const()[name = tensor("op_568_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116924992)))]; + tensor linear_26_cast_fp16 = linear(bias = var_568_to_fp16, weight = var_567_to_fp16, x = var_548_cast_fp16)[name = tensor("linear_26_cast_fp16")]; + tensor var_576 = const()[name = tensor("op_576"), val = tensor([1, 1500, 16, -1])]; + tensor var_577_cast_fp16 = reshape(shape = var_576, x = linear_24_cast_fp16)[name = tensor("op_577_cast_fp16")]; + tensor const_176_to_fp16 = const()[name = tensor("const_176_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast_fp16 = mul(x = var_577_cast_fp16, y = const_176_to_fp16)[name = tensor("q_19_cast_fp16")]; + tensor var_583 = const()[name = tensor("op_583"), val = tensor([1, 1500, 16, -1])]; + tensor var_584_cast_fp16 = reshape(shape = var_583, x = linear_25_cast_fp16)[name = tensor("op_584_cast_fp16")]; + tensor const_177_to_fp16 = const()[name = tensor("const_177_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast_fp16 = mul(x = var_584_cast_fp16, y = const_177_to_fp16)[name = tensor("k_19_cast_fp16")]; + tensor var_590 = const()[name = tensor("op_590"), val = tensor([1, 1500, 16, -1])]; + tensor var_591_cast_fp16 = reshape(shape = var_590, x = linear_26_cast_fp16)[name = tensor("op_591_cast_fp16")]; + tensor var_592 = const()[name = tensor("op_592"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_104_perm_0 = const()[name = tensor("transpose_104_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_105_perm_0 = const()[name = tensor("transpose_105_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_221 = transpose(perm = transpose_105_perm_0, x = k_19_cast_fp16)[name = tensor("transpose_221")]; + tensor transpose_222 = transpose(perm = transpose_104_perm_0, x = q_19_cast_fp16)[name = tensor("transpose_222")]; + tensor qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_222, y = transpose_221)[name = tensor("qk_9_cast_fp16")]; + tensor var_596_cast_fp16 = softmax(axis = var_532, x = qk_9_cast_fp16)[name = tensor("op_596_cast_fp16")]; + tensor var_598_transpose_x_0 = const()[name = tensor("op_598_transpose_x_0"), val = tensor(false)]; + tensor var_598_transpose_y_0 = const()[name = tensor("op_598_transpose_y_0"), val = tensor(false)]; + tensor transpose_223 = transpose(perm = var_592, x = var_591_cast_fp16)[name = tensor("transpose_223")]; + tensor var_598_cast_fp16 = matmul(transpose_x = var_598_transpose_x_0, transpose_y = var_598_transpose_y_0, x = var_596_cast_fp16, y = transpose_223)[name = tensor("op_598_cast_fp16")]; + tensor var_599 = const()[name = tensor("op_599"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 1024])]; + tensor transpose_220 = transpose(perm = var_599, x = var_598_cast_fp16)[name = tensor("transpose_220")]; + tensor x_59_cast_fp16 = reshape(shape = concat_4, x = transpose_220)[name = tensor("x_59_cast_fp16")]; + tensor var_604_to_fp16 = const()[name = tensor("op_604_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116927104)))]; + tensor var_605_to_fp16 = const()[name = tensor("op_605_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119024320)))]; + tensor linear_27_cast_fp16 = linear(bias = var_605_to_fp16, weight = var_604_to_fp16, x = x_59_cast_fp16)[name = tensor("linear_27_cast_fp16")]; + tensor x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = tensor("x_61_cast_fp16")]; + tensor var_612_axes_0 = const()[name = tensor("op_612_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119026432)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119028544)))]; + tensor var_612_cast_fp16 = layer_norm(axes = var_612_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_538_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = tensor("op_612_cast_fp16")]; + tensor var_621_to_fp16 = const()[name = tensor("op_621_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119030656)))]; + tensor var_622_to_fp16 = const()[name = tensor("op_622_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127419328)))]; + tensor linear_28_cast_fp16 = linear(bias = var_622_to_fp16, weight = var_621_to_fp16, x = var_612_cast_fp16)[name = tensor("linear_28_cast_fp16")]; + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = tensor("x_65_cast_fp16")]; + tensor var_627_to_fp16 = const()[name = tensor("op_627_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127427584)))]; + tensor var_628_to_fp16 = const()[name = tensor("op_628_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135816256)))]; + tensor linear_29_cast_fp16 = linear(bias = var_628_to_fp16, weight = var_627_to_fp16, x = x_65_cast_fp16)[name = tensor("linear_29_cast_fp16")]; + tensor x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = tensor("x_67_cast_fp16")]; + tensor var_638 = const()[name = tensor("op_638"), val = tensor(-1)]; + tensor var_654_axes_0 = const()[name = tensor("op_654_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135818368)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135820480)))]; + tensor var_644_to_fp16 = const()[name = tensor("op_644_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_654_cast_fp16 = layer_norm(axes = var_654_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_644_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = tensor("op_654_cast_fp16")]; + tensor var_665_to_fp16 = const()[name = tensor("op_665_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135822592)))]; + tensor var_666_to_fp16 = const()[name = tensor("op_666_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137919808)))]; + tensor linear_30_cast_fp16 = linear(bias = var_666_to_fp16, weight = var_665_to_fp16, x = var_654_cast_fp16)[name = tensor("linear_30_cast_fp16")]; + tensor var_669_to_fp16 = const()[name = tensor("op_669_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137921920)))]; + tensor linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_669_to_fp16, x = var_654_cast_fp16)[name = tensor("linear_31_cast_fp16")]; + tensor var_673_to_fp16 = const()[name = tensor("op_673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140019136)))]; + tensor var_674_to_fp16 = const()[name = tensor("op_674_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142116352)))]; + tensor linear_32_cast_fp16 = linear(bias = var_674_to_fp16, weight = var_673_to_fp16, x = var_654_cast_fp16)[name = tensor("linear_32_cast_fp16")]; + tensor var_682 = const()[name = tensor("op_682"), val = tensor([1, 1500, 16, -1])]; + tensor var_683_cast_fp16 = reshape(shape = var_682, x = linear_30_cast_fp16)[name = tensor("op_683_cast_fp16")]; + tensor const_178_to_fp16 = const()[name = tensor("const_178_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast_fp16 = mul(x = var_683_cast_fp16, y = const_178_to_fp16)[name = tensor("q_23_cast_fp16")]; + tensor var_689 = const()[name = tensor("op_689"), val = tensor([1, 1500, 16, -1])]; + tensor var_690_cast_fp16 = reshape(shape = var_689, x = linear_31_cast_fp16)[name = tensor("op_690_cast_fp16")]; + tensor const_179_to_fp16 = const()[name = tensor("const_179_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast_fp16 = mul(x = var_690_cast_fp16, y = const_179_to_fp16)[name = tensor("k_23_cast_fp16")]; + tensor var_696 = const()[name = tensor("op_696"), val = tensor([1, 1500, 16, -1])]; + tensor var_697_cast_fp16 = reshape(shape = var_696, x = linear_32_cast_fp16)[name = tensor("op_697_cast_fp16")]; + tensor var_698 = const()[name = tensor("op_698"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_106_perm_0 = const()[name = tensor("transpose_106_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_107_perm_0 = const()[name = tensor("transpose_107_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_217 = transpose(perm = transpose_107_perm_0, x = k_23_cast_fp16)[name = tensor("transpose_217")]; + tensor transpose_218 = transpose(perm = transpose_106_perm_0, x = q_23_cast_fp16)[name = tensor("transpose_218")]; + tensor qk_11_cast_fp16 = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_218, y = transpose_217)[name = tensor("qk_11_cast_fp16")]; + tensor var_702_cast_fp16 = softmax(axis = var_638, x = qk_11_cast_fp16)[name = tensor("op_702_cast_fp16")]; + tensor var_704_transpose_x_0 = const()[name = tensor("op_704_transpose_x_0"), val = tensor(false)]; + tensor var_704_transpose_y_0 = const()[name = tensor("op_704_transpose_y_0"), val = tensor(false)]; + tensor transpose_219 = transpose(perm = var_698, x = var_697_cast_fp16)[name = tensor("transpose_219")]; + tensor var_704_cast_fp16 = matmul(transpose_x = var_704_transpose_x_0, transpose_y = var_704_transpose_y_0, x = var_702_cast_fp16, y = transpose_219)[name = tensor("op_704_cast_fp16")]; + tensor var_705 = const()[name = tensor("op_705"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 1024])]; + tensor transpose_216 = transpose(perm = var_705, x = var_704_cast_fp16)[name = tensor("transpose_216")]; + tensor x_71_cast_fp16 = reshape(shape = concat_5, x = transpose_216)[name = tensor("x_71_cast_fp16")]; + tensor var_710_to_fp16 = const()[name = tensor("op_710_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142118464)))]; + tensor var_711_to_fp16 = const()[name = tensor("op_711_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144215680)))]; + tensor linear_33_cast_fp16 = linear(bias = var_711_to_fp16, weight = var_710_to_fp16, x = x_71_cast_fp16)[name = tensor("linear_33_cast_fp16")]; + tensor x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = tensor("x_73_cast_fp16")]; + tensor var_718_axes_0 = const()[name = tensor("op_718_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144217792)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144219904)))]; + tensor var_718_cast_fp16 = layer_norm(axes = var_718_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_644_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = tensor("op_718_cast_fp16")]; + tensor var_727_to_fp16 = const()[name = tensor("op_727_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144222016)))]; + tensor var_728_to_fp16 = const()[name = tensor("op_728_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152610688)))]; + tensor linear_34_cast_fp16 = linear(bias = var_728_to_fp16, weight = var_727_to_fp16, x = var_718_cast_fp16)[name = tensor("linear_34_cast_fp16")]; + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = tensor("x_77_cast_fp16")]; + tensor var_733_to_fp16 = const()[name = tensor("op_733_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152618944)))]; + tensor var_734_to_fp16 = const()[name = tensor("op_734_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161007616)))]; + tensor linear_35_cast_fp16 = linear(bias = var_734_to_fp16, weight = var_733_to_fp16, x = x_77_cast_fp16)[name = tensor("linear_35_cast_fp16")]; + tensor x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = tensor("x_79_cast_fp16")]; + tensor var_744 = const()[name = tensor("op_744"), val = tensor(-1)]; + tensor var_760_axes_0 = const()[name = tensor("op_760_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161009728)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161011840)))]; + tensor var_750_to_fp16 = const()[name = tensor("op_750_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_760_cast_fp16 = layer_norm(axes = var_760_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_750_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = tensor("op_760_cast_fp16")]; + tensor var_771_to_fp16 = const()[name = tensor("op_771_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161013952)))]; + tensor var_772_to_fp16 = const()[name = tensor("op_772_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163111168)))]; + tensor linear_36_cast_fp16 = linear(bias = var_772_to_fp16, weight = var_771_to_fp16, x = var_760_cast_fp16)[name = tensor("linear_36_cast_fp16")]; + tensor var_775_to_fp16 = const()[name = tensor("op_775_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163113280)))]; + tensor linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_775_to_fp16, x = var_760_cast_fp16)[name = tensor("linear_37_cast_fp16")]; + tensor var_779_to_fp16 = const()[name = tensor("op_779_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165210496)))]; + tensor var_780_to_fp16 = const()[name = tensor("op_780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167307712)))]; + tensor linear_38_cast_fp16 = linear(bias = var_780_to_fp16, weight = var_779_to_fp16, x = var_760_cast_fp16)[name = tensor("linear_38_cast_fp16")]; + tensor var_788 = const()[name = tensor("op_788"), val = tensor([1, 1500, 16, -1])]; + tensor var_789_cast_fp16 = reshape(shape = var_788, x = linear_36_cast_fp16)[name = tensor("op_789_cast_fp16")]; + tensor const_180_to_fp16 = const()[name = tensor("const_180_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast_fp16 = mul(x = var_789_cast_fp16, y = const_180_to_fp16)[name = tensor("q_27_cast_fp16")]; + tensor var_795 = const()[name = tensor("op_795"), val = tensor([1, 1500, 16, -1])]; + tensor var_796_cast_fp16 = reshape(shape = var_795, x = linear_37_cast_fp16)[name = tensor("op_796_cast_fp16")]; + tensor const_181_to_fp16 = const()[name = tensor("const_181_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast_fp16 = mul(x = var_796_cast_fp16, y = const_181_to_fp16)[name = tensor("k_27_cast_fp16")]; + tensor var_802 = const()[name = tensor("op_802"), val = tensor([1, 1500, 16, -1])]; + tensor var_803_cast_fp16 = reshape(shape = var_802, x = linear_38_cast_fp16)[name = tensor("op_803_cast_fp16")]; + tensor var_804 = const()[name = tensor("op_804"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_108_perm_0 = const()[name = tensor("transpose_108_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_109_perm_0 = const()[name = tensor("transpose_109_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_213 = transpose(perm = transpose_109_perm_0, x = k_27_cast_fp16)[name = tensor("transpose_213")]; + tensor transpose_214 = transpose(perm = transpose_108_perm_0, x = q_27_cast_fp16)[name = tensor("transpose_214")]; + tensor qk_13_cast_fp16 = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_214, y = transpose_213)[name = tensor("qk_13_cast_fp16")]; + tensor var_808_cast_fp16 = softmax(axis = var_744, x = qk_13_cast_fp16)[name = tensor("op_808_cast_fp16")]; + tensor var_810_transpose_x_0 = const()[name = tensor("op_810_transpose_x_0"), val = tensor(false)]; + tensor var_810_transpose_y_0 = const()[name = tensor("op_810_transpose_y_0"), val = tensor(false)]; + tensor transpose_215 = transpose(perm = var_804, x = var_803_cast_fp16)[name = tensor("transpose_215")]; + tensor var_810_cast_fp16 = matmul(transpose_x = var_810_transpose_x_0, transpose_y = var_810_transpose_y_0, x = var_808_cast_fp16, y = transpose_215)[name = tensor("op_810_cast_fp16")]; + tensor var_811 = const()[name = tensor("op_811"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 1024])]; + tensor transpose_212 = transpose(perm = var_811, x = var_810_cast_fp16)[name = tensor("transpose_212")]; + tensor x_83_cast_fp16 = reshape(shape = concat_6, x = transpose_212)[name = tensor("x_83_cast_fp16")]; + tensor var_816_to_fp16 = const()[name = tensor("op_816_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167309824)))]; + tensor var_817_to_fp16 = const()[name = tensor("op_817_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169407040)))]; + tensor linear_39_cast_fp16 = linear(bias = var_817_to_fp16, weight = var_816_to_fp16, x = x_83_cast_fp16)[name = tensor("linear_39_cast_fp16")]; + tensor x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = tensor("x_85_cast_fp16")]; + tensor var_824_axes_0 = const()[name = tensor("op_824_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169409152)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169411264)))]; + tensor var_824_cast_fp16 = layer_norm(axes = var_824_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_750_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = tensor("op_824_cast_fp16")]; + tensor var_833_to_fp16 = const()[name = tensor("op_833_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169413376)))]; + tensor var_834_to_fp16 = const()[name = tensor("op_834_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177802048)))]; + tensor linear_40_cast_fp16 = linear(bias = var_834_to_fp16, weight = var_833_to_fp16, x = var_824_cast_fp16)[name = tensor("linear_40_cast_fp16")]; + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = tensor("x_89_cast_fp16")]; + tensor var_839_to_fp16 = const()[name = tensor("op_839_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177810304)))]; + tensor var_840_to_fp16 = const()[name = tensor("op_840_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186198976)))]; + tensor linear_41_cast_fp16 = linear(bias = var_840_to_fp16, weight = var_839_to_fp16, x = x_89_cast_fp16)[name = tensor("linear_41_cast_fp16")]; + tensor x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = tensor("x_91_cast_fp16")]; + tensor var_850 = const()[name = tensor("op_850"), val = tensor(-1)]; + tensor var_866_axes_0 = const()[name = tensor("op_866_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186201088)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186203200)))]; + tensor var_856_to_fp16 = const()[name = tensor("op_856_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_866_cast_fp16 = layer_norm(axes = var_866_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_856_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = tensor("op_866_cast_fp16")]; + tensor var_877_to_fp16 = const()[name = tensor("op_877_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186205312)))]; + tensor var_878_to_fp16 = const()[name = tensor("op_878_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188302528)))]; + tensor linear_42_cast_fp16 = linear(bias = var_878_to_fp16, weight = var_877_to_fp16, x = var_866_cast_fp16)[name = tensor("linear_42_cast_fp16")]; + tensor var_881_to_fp16 = const()[name = tensor("op_881_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188304640)))]; + tensor linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_881_to_fp16, x = var_866_cast_fp16)[name = tensor("linear_43_cast_fp16")]; + tensor var_885_to_fp16 = const()[name = tensor("op_885_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190401856)))]; + tensor var_886_to_fp16 = const()[name = tensor("op_886_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192499072)))]; + tensor linear_44_cast_fp16 = linear(bias = var_886_to_fp16, weight = var_885_to_fp16, x = var_866_cast_fp16)[name = tensor("linear_44_cast_fp16")]; + tensor var_894 = const()[name = tensor("op_894"), val = tensor([1, 1500, 16, -1])]; + tensor var_895_cast_fp16 = reshape(shape = var_894, x = linear_42_cast_fp16)[name = tensor("op_895_cast_fp16")]; + tensor const_182_to_fp16 = const()[name = tensor("const_182_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast_fp16 = mul(x = var_895_cast_fp16, y = const_182_to_fp16)[name = tensor("q_31_cast_fp16")]; + tensor var_901 = const()[name = tensor("op_901"), val = tensor([1, 1500, 16, -1])]; + tensor var_902_cast_fp16 = reshape(shape = var_901, x = linear_43_cast_fp16)[name = tensor("op_902_cast_fp16")]; + tensor const_183_to_fp16 = const()[name = tensor("const_183_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast_fp16 = mul(x = var_902_cast_fp16, y = const_183_to_fp16)[name = tensor("k_31_cast_fp16")]; + tensor var_908 = const()[name = tensor("op_908"), val = tensor([1, 1500, 16, -1])]; + tensor var_909_cast_fp16 = reshape(shape = var_908, x = linear_44_cast_fp16)[name = tensor("op_909_cast_fp16")]; + tensor var_910 = const()[name = tensor("op_910"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_110_perm_0 = const()[name = tensor("transpose_110_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_111_perm_0 = const()[name = tensor("transpose_111_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_209 = transpose(perm = transpose_111_perm_0, x = k_31_cast_fp16)[name = tensor("transpose_209")]; + tensor transpose_210 = transpose(perm = transpose_110_perm_0, x = q_31_cast_fp16)[name = tensor("transpose_210")]; + tensor qk_15_cast_fp16 = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_210, y = transpose_209)[name = tensor("qk_15_cast_fp16")]; + tensor var_914_cast_fp16 = softmax(axis = var_850, x = qk_15_cast_fp16)[name = tensor("op_914_cast_fp16")]; + tensor var_916_transpose_x_0 = const()[name = tensor("op_916_transpose_x_0"), val = tensor(false)]; + tensor var_916_transpose_y_0 = const()[name = tensor("op_916_transpose_y_0"), val = tensor(false)]; + tensor transpose_211 = transpose(perm = var_910, x = var_909_cast_fp16)[name = tensor("transpose_211")]; + tensor var_916_cast_fp16 = matmul(transpose_x = var_916_transpose_x_0, transpose_y = var_916_transpose_y_0, x = var_914_cast_fp16, y = transpose_211)[name = tensor("op_916_cast_fp16")]; + tensor var_917 = const()[name = tensor("op_917"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 1024])]; + tensor transpose_208 = transpose(perm = var_917, x = var_916_cast_fp16)[name = tensor("transpose_208")]; + tensor x_95_cast_fp16 = reshape(shape = concat_7, x = transpose_208)[name = tensor("x_95_cast_fp16")]; + tensor var_922_to_fp16 = const()[name = tensor("op_922_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192501184)))]; + tensor var_923_to_fp16 = const()[name = tensor("op_923_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194598400)))]; + tensor linear_45_cast_fp16 = linear(bias = var_923_to_fp16, weight = var_922_to_fp16, x = x_95_cast_fp16)[name = tensor("linear_45_cast_fp16")]; + tensor x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = tensor("x_97_cast_fp16")]; + tensor var_930_axes_0 = const()[name = tensor("op_930_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194600512)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194602624)))]; + tensor var_930_cast_fp16 = layer_norm(axes = var_930_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_856_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = tensor("op_930_cast_fp16")]; + tensor var_939_to_fp16 = const()[name = tensor("op_939_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194604736)))]; + tensor var_940_to_fp16 = const()[name = tensor("op_940_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202993408)))]; + tensor linear_46_cast_fp16 = linear(bias = var_940_to_fp16, weight = var_939_to_fp16, x = var_930_cast_fp16)[name = tensor("linear_46_cast_fp16")]; + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = tensor("x_101_cast_fp16")]; + tensor var_945_to_fp16 = const()[name = tensor("op_945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203001664)))]; + tensor var_946_to_fp16 = const()[name = tensor("op_946_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211390336)))]; + tensor linear_47_cast_fp16 = linear(bias = var_946_to_fp16, weight = var_945_to_fp16, x = x_101_cast_fp16)[name = tensor("linear_47_cast_fp16")]; + tensor x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = tensor("x_103_cast_fp16")]; + tensor var_956 = const()[name = tensor("op_956"), val = tensor(-1)]; + tensor var_972_axes_0 = const()[name = tensor("op_972_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211392448)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211394560)))]; + tensor var_962_to_fp16 = const()[name = tensor("op_962_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_972_cast_fp16 = layer_norm(axes = var_972_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_962_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = tensor("op_972_cast_fp16")]; + tensor var_983_to_fp16 = const()[name = tensor("op_983_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211396672)))]; + tensor var_984_to_fp16 = const()[name = tensor("op_984_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213493888)))]; + tensor linear_48_cast_fp16 = linear(bias = var_984_to_fp16, weight = var_983_to_fp16, x = var_972_cast_fp16)[name = tensor("linear_48_cast_fp16")]; + tensor var_987_to_fp16 = const()[name = tensor("op_987_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213496000)))]; + tensor linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_987_to_fp16, x = var_972_cast_fp16)[name = tensor("linear_49_cast_fp16")]; + tensor var_991_to_fp16 = const()[name = tensor("op_991_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215593216)))]; + tensor var_992_to_fp16 = const()[name = tensor("op_992_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217690432)))]; + tensor linear_50_cast_fp16 = linear(bias = var_992_to_fp16, weight = var_991_to_fp16, x = var_972_cast_fp16)[name = tensor("linear_50_cast_fp16")]; + tensor var_1000 = const()[name = tensor("op_1000"), val = tensor([1, 1500, 16, -1])]; + tensor var_1001_cast_fp16 = reshape(shape = var_1000, x = linear_48_cast_fp16)[name = tensor("op_1001_cast_fp16")]; + tensor const_184_to_fp16 = const()[name = tensor("const_184_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast_fp16 = mul(x = var_1001_cast_fp16, y = const_184_to_fp16)[name = tensor("q_35_cast_fp16")]; + tensor var_1007 = const()[name = tensor("op_1007"), val = tensor([1, 1500, 16, -1])]; + tensor var_1008_cast_fp16 = reshape(shape = var_1007, x = linear_49_cast_fp16)[name = tensor("op_1008_cast_fp16")]; + tensor const_185_to_fp16 = const()[name = tensor("const_185_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast_fp16 = mul(x = var_1008_cast_fp16, y = const_185_to_fp16)[name = tensor("k_35_cast_fp16")]; + tensor var_1014 = const()[name = tensor("op_1014"), val = tensor([1, 1500, 16, -1])]; + tensor var_1015_cast_fp16 = reshape(shape = var_1014, x = linear_50_cast_fp16)[name = tensor("op_1015_cast_fp16")]; + tensor var_1016 = const()[name = tensor("op_1016"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_112_perm_0 = const()[name = tensor("transpose_112_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_113_perm_0 = const()[name = tensor("transpose_113_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_205 = transpose(perm = transpose_113_perm_0, x = k_35_cast_fp16)[name = tensor("transpose_205")]; + tensor transpose_206 = transpose(perm = transpose_112_perm_0, x = q_35_cast_fp16)[name = tensor("transpose_206")]; + tensor qk_17_cast_fp16 = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_206, y = transpose_205)[name = tensor("qk_17_cast_fp16")]; + tensor var_1020_cast_fp16 = softmax(axis = var_956, x = qk_17_cast_fp16)[name = tensor("op_1020_cast_fp16")]; + tensor var_1022_transpose_x_0 = const()[name = tensor("op_1022_transpose_x_0"), val = tensor(false)]; + tensor var_1022_transpose_y_0 = const()[name = tensor("op_1022_transpose_y_0"), val = tensor(false)]; + tensor transpose_207 = transpose(perm = var_1016, x = var_1015_cast_fp16)[name = tensor("transpose_207")]; + tensor var_1022_cast_fp16 = matmul(transpose_x = var_1022_transpose_x_0, transpose_y = var_1022_transpose_y_0, x = var_1020_cast_fp16, y = transpose_207)[name = tensor("op_1022_cast_fp16")]; + tensor var_1023 = const()[name = tensor("op_1023"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 1024])]; + tensor transpose_204 = transpose(perm = var_1023, x = var_1022_cast_fp16)[name = tensor("transpose_204")]; + tensor x_107_cast_fp16 = reshape(shape = concat_8, x = transpose_204)[name = tensor("x_107_cast_fp16")]; + tensor var_1028_to_fp16 = const()[name = tensor("op_1028_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217692544)))]; + tensor var_1029_to_fp16 = const()[name = tensor("op_1029_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219789760)))]; + tensor linear_51_cast_fp16 = linear(bias = var_1029_to_fp16, weight = var_1028_to_fp16, x = x_107_cast_fp16)[name = tensor("linear_51_cast_fp16")]; + tensor x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = tensor("x_109_cast_fp16")]; + tensor var_1036_axes_0 = const()[name = tensor("op_1036_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219791872)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219793984)))]; + tensor var_1036_cast_fp16 = layer_norm(axes = var_1036_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_962_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = tensor("op_1036_cast_fp16")]; + tensor var_1045_to_fp16 = const()[name = tensor("op_1045_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219796096)))]; + tensor var_1046_to_fp16 = const()[name = tensor("op_1046_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228184768)))]; + tensor linear_52_cast_fp16 = linear(bias = var_1046_to_fp16, weight = var_1045_to_fp16, x = var_1036_cast_fp16)[name = tensor("linear_52_cast_fp16")]; + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = tensor("x_113_cast_fp16")]; + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228193024)))]; + tensor var_1052_to_fp16 = const()[name = tensor("op_1052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236581696)))]; + tensor linear_53_cast_fp16 = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_113_cast_fp16)[name = tensor("linear_53_cast_fp16")]; + tensor x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = tensor("x_115_cast_fp16")]; + tensor var_1062 = const()[name = tensor("op_1062"), val = tensor(-1)]; + tensor var_1078_axes_0 = const()[name = tensor("op_1078_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236583808)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236585920)))]; + tensor var_1068_to_fp16 = const()[name = tensor("op_1068_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1078_cast_fp16 = layer_norm(axes = var_1078_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1068_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = tensor("op_1078_cast_fp16")]; + tensor var_1089_to_fp16 = const()[name = tensor("op_1089_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236588032)))]; + tensor var_1090_to_fp16 = const()[name = tensor("op_1090_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238685248)))]; + tensor linear_54_cast_fp16 = linear(bias = var_1090_to_fp16, weight = var_1089_to_fp16, x = var_1078_cast_fp16)[name = tensor("linear_54_cast_fp16")]; + tensor var_1093_to_fp16 = const()[name = tensor("op_1093_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238687360)))]; + tensor linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1093_to_fp16, x = var_1078_cast_fp16)[name = tensor("linear_55_cast_fp16")]; + tensor var_1097_to_fp16 = const()[name = tensor("op_1097_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240784576)))]; + tensor var_1098_to_fp16 = const()[name = tensor("op_1098_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242881792)))]; + tensor linear_56_cast_fp16 = linear(bias = var_1098_to_fp16, weight = var_1097_to_fp16, x = var_1078_cast_fp16)[name = tensor("linear_56_cast_fp16")]; + tensor var_1106 = const()[name = tensor("op_1106"), val = tensor([1, 1500, 16, -1])]; + tensor var_1107_cast_fp16 = reshape(shape = var_1106, x = linear_54_cast_fp16)[name = tensor("op_1107_cast_fp16")]; + tensor const_186_to_fp16 = const()[name = tensor("const_186_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast_fp16 = mul(x = var_1107_cast_fp16, y = const_186_to_fp16)[name = tensor("q_39_cast_fp16")]; + tensor var_1113 = const()[name = tensor("op_1113"), val = tensor([1, 1500, 16, -1])]; + tensor var_1114_cast_fp16 = reshape(shape = var_1113, x = linear_55_cast_fp16)[name = tensor("op_1114_cast_fp16")]; + tensor const_187_to_fp16 = const()[name = tensor("const_187_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast_fp16 = mul(x = var_1114_cast_fp16, y = const_187_to_fp16)[name = tensor("k_39_cast_fp16")]; + tensor var_1120 = const()[name = tensor("op_1120"), val = tensor([1, 1500, 16, -1])]; + tensor var_1121_cast_fp16 = reshape(shape = var_1120, x = linear_56_cast_fp16)[name = tensor("op_1121_cast_fp16")]; + tensor var_1122 = const()[name = tensor("op_1122"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_114_perm_0 = const()[name = tensor("transpose_114_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_115_perm_0 = const()[name = tensor("transpose_115_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_201 = transpose(perm = transpose_115_perm_0, x = k_39_cast_fp16)[name = tensor("transpose_201")]; + tensor transpose_202 = transpose(perm = transpose_114_perm_0, x = q_39_cast_fp16)[name = tensor("transpose_202")]; + tensor qk_19_cast_fp16 = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_202, y = transpose_201)[name = tensor("qk_19_cast_fp16")]; + tensor var_1126_cast_fp16 = softmax(axis = var_1062, x = qk_19_cast_fp16)[name = tensor("op_1126_cast_fp16")]; + tensor var_1128_transpose_x_0 = const()[name = tensor("op_1128_transpose_x_0"), val = tensor(false)]; + tensor var_1128_transpose_y_0 = const()[name = tensor("op_1128_transpose_y_0"), val = tensor(false)]; + tensor transpose_203 = transpose(perm = var_1122, x = var_1121_cast_fp16)[name = tensor("transpose_203")]; + tensor var_1128_cast_fp16 = matmul(transpose_x = var_1128_transpose_x_0, transpose_y = var_1128_transpose_y_0, x = var_1126_cast_fp16, y = transpose_203)[name = tensor("op_1128_cast_fp16")]; + tensor var_1129 = const()[name = tensor("op_1129"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 1024])]; + tensor transpose_200 = transpose(perm = var_1129, x = var_1128_cast_fp16)[name = tensor("transpose_200")]; + tensor x_119_cast_fp16 = reshape(shape = concat_9, x = transpose_200)[name = tensor("x_119_cast_fp16")]; + tensor var_1134_to_fp16 = const()[name = tensor("op_1134_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242883904)))]; + tensor var_1135_to_fp16 = const()[name = tensor("op_1135_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244981120)))]; + tensor linear_57_cast_fp16 = linear(bias = var_1135_to_fp16, weight = var_1134_to_fp16, x = x_119_cast_fp16)[name = tensor("linear_57_cast_fp16")]; + tensor x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = tensor("x_121_cast_fp16")]; + tensor var_1142_axes_0 = const()[name = tensor("op_1142_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244983232)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244985344)))]; + tensor var_1142_cast_fp16 = layer_norm(axes = var_1142_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1068_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = tensor("op_1142_cast_fp16")]; + tensor var_1151_to_fp16 = const()[name = tensor("op_1151_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244987456)))]; + tensor var_1152_to_fp16 = const()[name = tensor("op_1152_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253376128)))]; + tensor linear_58_cast_fp16 = linear(bias = var_1152_to_fp16, weight = var_1151_to_fp16, x = var_1142_cast_fp16)[name = tensor("linear_58_cast_fp16")]; + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = tensor("x_125_cast_fp16")]; + tensor var_1157_to_fp16 = const()[name = tensor("op_1157_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253384384)))]; + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261773056)))]; + tensor linear_59_cast_fp16 = linear(bias = var_1158_to_fp16, weight = var_1157_to_fp16, x = x_125_cast_fp16)[name = tensor("linear_59_cast_fp16")]; + tensor x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = tensor("x_127_cast_fp16")]; + tensor var_1168 = const()[name = tensor("op_1168"), val = tensor(-1)]; + tensor var_1184_axes_0 = const()[name = tensor("op_1184_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261775168)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261777280)))]; + tensor var_1174_to_fp16 = const()[name = tensor("op_1174_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1184_cast_fp16 = layer_norm(axes = var_1184_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1174_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = tensor("op_1184_cast_fp16")]; + tensor var_1195_to_fp16 = const()[name = tensor("op_1195_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261779392)))]; + tensor var_1196_to_fp16 = const()[name = tensor("op_1196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263876608)))]; + tensor linear_60_cast_fp16 = linear(bias = var_1196_to_fp16, weight = var_1195_to_fp16, x = var_1184_cast_fp16)[name = tensor("linear_60_cast_fp16")]; + tensor var_1199_to_fp16 = const()[name = tensor("op_1199_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263878720)))]; + tensor linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1199_to_fp16, x = var_1184_cast_fp16)[name = tensor("linear_61_cast_fp16")]; + tensor var_1203_to_fp16 = const()[name = tensor("op_1203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265975936)))]; + tensor var_1204_to_fp16 = const()[name = tensor("op_1204_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268073152)))]; + tensor linear_62_cast_fp16 = linear(bias = var_1204_to_fp16, weight = var_1203_to_fp16, x = var_1184_cast_fp16)[name = tensor("linear_62_cast_fp16")]; + tensor var_1212 = const()[name = tensor("op_1212"), val = tensor([1, 1500, 16, -1])]; + tensor var_1213_cast_fp16 = reshape(shape = var_1212, x = linear_60_cast_fp16)[name = tensor("op_1213_cast_fp16")]; + tensor const_188_to_fp16 = const()[name = tensor("const_188_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast_fp16 = mul(x = var_1213_cast_fp16, y = const_188_to_fp16)[name = tensor("q_43_cast_fp16")]; + tensor var_1219 = const()[name = tensor("op_1219"), val = tensor([1, 1500, 16, -1])]; + tensor var_1220_cast_fp16 = reshape(shape = var_1219, x = linear_61_cast_fp16)[name = tensor("op_1220_cast_fp16")]; + tensor const_189_to_fp16 = const()[name = tensor("const_189_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast_fp16 = mul(x = var_1220_cast_fp16, y = const_189_to_fp16)[name = tensor("k_43_cast_fp16")]; + tensor var_1226 = const()[name = tensor("op_1226"), val = tensor([1, 1500, 16, -1])]; + tensor var_1227_cast_fp16 = reshape(shape = var_1226, x = linear_62_cast_fp16)[name = tensor("op_1227_cast_fp16")]; + tensor var_1228 = const()[name = tensor("op_1228"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_116_perm_0 = const()[name = tensor("transpose_116_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_117_perm_0 = const()[name = tensor("transpose_117_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_197 = transpose(perm = transpose_117_perm_0, x = k_43_cast_fp16)[name = tensor("transpose_197")]; + tensor transpose_198 = transpose(perm = transpose_116_perm_0, x = q_43_cast_fp16)[name = tensor("transpose_198")]; + tensor qk_21_cast_fp16 = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_198, y = transpose_197)[name = tensor("qk_21_cast_fp16")]; + tensor var_1232_cast_fp16 = softmax(axis = var_1168, x = qk_21_cast_fp16)[name = tensor("op_1232_cast_fp16")]; + tensor var_1234_transpose_x_0 = const()[name = tensor("op_1234_transpose_x_0"), val = tensor(false)]; + tensor var_1234_transpose_y_0 = const()[name = tensor("op_1234_transpose_y_0"), val = tensor(false)]; + tensor transpose_199 = transpose(perm = var_1228, x = var_1227_cast_fp16)[name = tensor("transpose_199")]; + tensor var_1234_cast_fp16 = matmul(transpose_x = var_1234_transpose_x_0, transpose_y = var_1234_transpose_y_0, x = var_1232_cast_fp16, y = transpose_199)[name = tensor("op_1234_cast_fp16")]; + tensor var_1235 = const()[name = tensor("op_1235"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 1024])]; + tensor transpose_196 = transpose(perm = var_1235, x = var_1234_cast_fp16)[name = tensor("transpose_196")]; + tensor x_131_cast_fp16 = reshape(shape = concat_10, x = transpose_196)[name = tensor("x_131_cast_fp16")]; + tensor var_1240_to_fp16 = const()[name = tensor("op_1240_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268075264)))]; + tensor var_1241_to_fp16 = const()[name = tensor("op_1241_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270172480)))]; + tensor linear_63_cast_fp16 = linear(bias = var_1241_to_fp16, weight = var_1240_to_fp16, x = x_131_cast_fp16)[name = tensor("linear_63_cast_fp16")]; + tensor x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = tensor("x_133_cast_fp16")]; + tensor var_1248_axes_0 = const()[name = tensor("op_1248_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270174592)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270176704)))]; + tensor var_1248_cast_fp16 = layer_norm(axes = var_1248_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1174_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = tensor("op_1248_cast_fp16")]; + tensor var_1257_to_fp16 = const()[name = tensor("op_1257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270178816)))]; + tensor var_1258_to_fp16 = const()[name = tensor("op_1258_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278567488)))]; + tensor linear_64_cast_fp16 = linear(bias = var_1258_to_fp16, weight = var_1257_to_fp16, x = var_1248_cast_fp16)[name = tensor("linear_64_cast_fp16")]; + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = tensor("x_137_cast_fp16")]; + tensor var_1263_to_fp16 = const()[name = tensor("op_1263_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278575744)))]; + tensor var_1264_to_fp16 = const()[name = tensor("op_1264_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286964416)))]; + tensor linear_65_cast_fp16 = linear(bias = var_1264_to_fp16, weight = var_1263_to_fp16, x = x_137_cast_fp16)[name = tensor("linear_65_cast_fp16")]; + tensor x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = tensor("x_139_cast_fp16")]; + tensor var_1274 = const()[name = tensor("op_1274"), val = tensor(-1)]; + tensor var_1290_axes_0 = const()[name = tensor("op_1290_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286966528)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286968640)))]; + tensor var_1280_to_fp16 = const()[name = tensor("op_1280_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1290_cast_fp16 = layer_norm(axes = var_1290_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1280_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = tensor("op_1290_cast_fp16")]; + tensor var_1301_to_fp16 = const()[name = tensor("op_1301_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286970752)))]; + tensor var_1302_to_fp16 = const()[name = tensor("op_1302_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289067968)))]; + tensor linear_66_cast_fp16 = linear(bias = var_1302_to_fp16, weight = var_1301_to_fp16, x = var_1290_cast_fp16)[name = tensor("linear_66_cast_fp16")]; + tensor var_1305_to_fp16 = const()[name = tensor("op_1305_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289070080)))]; + tensor linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1305_to_fp16, x = var_1290_cast_fp16)[name = tensor("linear_67_cast_fp16")]; + tensor var_1309_to_fp16 = const()[name = tensor("op_1309_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291167296)))]; + tensor var_1310_to_fp16 = const()[name = tensor("op_1310_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293264512)))]; + tensor linear_68_cast_fp16 = linear(bias = var_1310_to_fp16, weight = var_1309_to_fp16, x = var_1290_cast_fp16)[name = tensor("linear_68_cast_fp16")]; + tensor var_1318 = const()[name = tensor("op_1318"), val = tensor([1, 1500, 16, -1])]; + tensor var_1319_cast_fp16 = reshape(shape = var_1318, x = linear_66_cast_fp16)[name = tensor("op_1319_cast_fp16")]; + tensor const_190_to_fp16 = const()[name = tensor("const_190_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast_fp16 = mul(x = var_1319_cast_fp16, y = const_190_to_fp16)[name = tensor("q_47_cast_fp16")]; + tensor var_1325 = const()[name = tensor("op_1325"), val = tensor([1, 1500, 16, -1])]; + tensor var_1326_cast_fp16 = reshape(shape = var_1325, x = linear_67_cast_fp16)[name = tensor("op_1326_cast_fp16")]; + tensor const_191_to_fp16 = const()[name = tensor("const_191_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast_fp16 = mul(x = var_1326_cast_fp16, y = const_191_to_fp16)[name = tensor("k_47_cast_fp16")]; + tensor var_1332 = const()[name = tensor("op_1332"), val = tensor([1, 1500, 16, -1])]; + tensor var_1333_cast_fp16 = reshape(shape = var_1332, x = linear_68_cast_fp16)[name = tensor("op_1333_cast_fp16")]; + tensor var_1334 = const()[name = tensor("op_1334"), val = tensor([0, 2, 1, 3])]; + tensor qk_23_transpose_x_0 = const()[name = tensor("qk_23_transpose_x_0"), val = tensor(false)]; + tensor qk_23_transpose_y_0 = const()[name = tensor("qk_23_transpose_y_0"), val = tensor(false)]; + tensor transpose_118_perm_0 = const()[name = tensor("transpose_118_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_119_perm_0 = const()[name = tensor("transpose_119_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_193 = transpose(perm = transpose_119_perm_0, x = k_47_cast_fp16)[name = tensor("transpose_193")]; + tensor transpose_194 = transpose(perm = transpose_118_perm_0, x = q_47_cast_fp16)[name = tensor("transpose_194")]; + tensor qk_23_cast_fp16 = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_194, y = transpose_193)[name = tensor("qk_23_cast_fp16")]; + tensor var_1338_cast_fp16 = softmax(axis = var_1274, x = qk_23_cast_fp16)[name = tensor("op_1338_cast_fp16")]; + tensor var_1340_transpose_x_0 = const()[name = tensor("op_1340_transpose_x_0"), val = tensor(false)]; + tensor var_1340_transpose_y_0 = const()[name = tensor("op_1340_transpose_y_0"), val = tensor(false)]; + tensor transpose_195 = transpose(perm = var_1334, x = var_1333_cast_fp16)[name = tensor("transpose_195")]; + tensor var_1340_cast_fp16 = matmul(transpose_x = var_1340_transpose_x_0, transpose_y = var_1340_transpose_y_0, x = var_1338_cast_fp16, y = transpose_195)[name = tensor("op_1340_cast_fp16")]; + tensor var_1341 = const()[name = tensor("op_1341"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 1024])]; + tensor transpose_192 = transpose(perm = var_1341, x = var_1340_cast_fp16)[name = tensor("transpose_192")]; + tensor x_143_cast_fp16 = reshape(shape = concat_11, x = transpose_192)[name = tensor("x_143_cast_fp16")]; + tensor var_1346_to_fp16 = const()[name = tensor("op_1346_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293266624)))]; + tensor var_1347_to_fp16 = const()[name = tensor("op_1347_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295363840)))]; + tensor linear_69_cast_fp16 = linear(bias = var_1347_to_fp16, weight = var_1346_to_fp16, x = x_143_cast_fp16)[name = tensor("linear_69_cast_fp16")]; + tensor x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = tensor("x_145_cast_fp16")]; + tensor var_1354_axes_0 = const()[name = tensor("op_1354_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295365952)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295368064)))]; + tensor var_1354_cast_fp16 = layer_norm(axes = var_1354_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1280_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = tensor("op_1354_cast_fp16")]; + tensor var_1363_to_fp16 = const()[name = tensor("op_1363_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295370176)))]; + tensor var_1364_to_fp16 = const()[name = tensor("op_1364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303758848)))]; + tensor linear_70_cast_fp16 = linear(bias = var_1364_to_fp16, weight = var_1363_to_fp16, x = var_1354_cast_fp16)[name = tensor("linear_70_cast_fp16")]; + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = tensor("x_149_cast_fp16")]; + tensor var_1369_to_fp16 = const()[name = tensor("op_1369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303767104)))]; + tensor var_1370_to_fp16 = const()[name = tensor("op_1370_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312155776)))]; + tensor linear_71_cast_fp16 = linear(bias = var_1370_to_fp16, weight = var_1369_to_fp16, x = x_149_cast_fp16)[name = tensor("linear_71_cast_fp16")]; + tensor x_151_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = tensor("x_151_cast_fp16")]; + tensor var_1380 = const()[name = tensor("op_1380"), val = tensor(-1)]; + tensor var_1396_axes_0 = const()[name = tensor("op_1396_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312157888)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312160000)))]; + tensor var_1386_to_fp16 = const()[name = tensor("op_1386_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1396_cast_fp16 = layer_norm(axes = var_1396_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1386_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = tensor("op_1396_cast_fp16")]; + tensor var_1407_to_fp16 = const()[name = tensor("op_1407_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312162112)))]; + tensor var_1408_to_fp16 = const()[name = tensor("op_1408_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314259328)))]; + tensor linear_72_cast_fp16 = linear(bias = var_1408_to_fp16, weight = var_1407_to_fp16, x = var_1396_cast_fp16)[name = tensor("linear_72_cast_fp16")]; + tensor var_1411_to_fp16 = const()[name = tensor("op_1411_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314261440)))]; + tensor linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1411_to_fp16, x = var_1396_cast_fp16)[name = tensor("linear_73_cast_fp16")]; + tensor var_1415_to_fp16 = const()[name = tensor("op_1415_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316358656)))]; + tensor var_1416_to_fp16 = const()[name = tensor("op_1416_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318455872)))]; + tensor linear_74_cast_fp16 = linear(bias = var_1416_to_fp16, weight = var_1415_to_fp16, x = var_1396_cast_fp16)[name = tensor("linear_74_cast_fp16")]; + tensor var_1424 = const()[name = tensor("op_1424"), val = tensor([1, 1500, 16, -1])]; + tensor var_1425_cast_fp16 = reshape(shape = var_1424, x = linear_72_cast_fp16)[name = tensor("op_1425_cast_fp16")]; + tensor const_192_to_fp16 = const()[name = tensor("const_192_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast_fp16 = mul(x = var_1425_cast_fp16, y = const_192_to_fp16)[name = tensor("q_51_cast_fp16")]; + tensor var_1431 = const()[name = tensor("op_1431"), val = tensor([1, 1500, 16, -1])]; + tensor var_1432_cast_fp16 = reshape(shape = var_1431, x = linear_73_cast_fp16)[name = tensor("op_1432_cast_fp16")]; + tensor const_193_to_fp16 = const()[name = tensor("const_193_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast_fp16 = mul(x = var_1432_cast_fp16, y = const_193_to_fp16)[name = tensor("k_51_cast_fp16")]; + tensor var_1438 = const()[name = tensor("op_1438"), val = tensor([1, 1500, 16, -1])]; + tensor var_1439_cast_fp16 = reshape(shape = var_1438, x = linear_74_cast_fp16)[name = tensor("op_1439_cast_fp16")]; + tensor var_1440 = const()[name = tensor("op_1440"), val = tensor([0, 2, 1, 3])]; + tensor qk_25_transpose_x_0 = const()[name = tensor("qk_25_transpose_x_0"), val = tensor(false)]; + tensor qk_25_transpose_y_0 = const()[name = tensor("qk_25_transpose_y_0"), val = tensor(false)]; + tensor transpose_120_perm_0 = const()[name = tensor("transpose_120_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_121_perm_0 = const()[name = tensor("transpose_121_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_189 = transpose(perm = transpose_121_perm_0, x = k_51_cast_fp16)[name = tensor("transpose_189")]; + tensor transpose_190 = transpose(perm = transpose_120_perm_0, x = q_51_cast_fp16)[name = tensor("transpose_190")]; + tensor qk_25_cast_fp16 = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_190, y = transpose_189)[name = tensor("qk_25_cast_fp16")]; + tensor var_1444_cast_fp16 = softmax(axis = var_1380, x = qk_25_cast_fp16)[name = tensor("op_1444_cast_fp16")]; + tensor var_1446_transpose_x_0 = const()[name = tensor("op_1446_transpose_x_0"), val = tensor(false)]; + tensor var_1446_transpose_y_0 = const()[name = tensor("op_1446_transpose_y_0"), val = tensor(false)]; + tensor transpose_191 = transpose(perm = var_1440, x = var_1439_cast_fp16)[name = tensor("transpose_191")]; + tensor var_1446_cast_fp16 = matmul(transpose_x = var_1446_transpose_x_0, transpose_y = var_1446_transpose_y_0, x = var_1444_cast_fp16, y = transpose_191)[name = tensor("op_1446_cast_fp16")]; + tensor var_1447 = const()[name = tensor("op_1447"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = tensor("concat_12"), val = tensor([1, 1500, 1024])]; + tensor transpose_188 = transpose(perm = var_1447, x = var_1446_cast_fp16)[name = tensor("transpose_188")]; + tensor x_155_cast_fp16 = reshape(shape = concat_12, x = transpose_188)[name = tensor("x_155_cast_fp16")]; + tensor var_1452_to_fp16 = const()[name = tensor("op_1452_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318457984)))]; + tensor var_1453_to_fp16 = const()[name = tensor("op_1453_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320555200)))]; + tensor linear_75_cast_fp16 = linear(bias = var_1453_to_fp16, weight = var_1452_to_fp16, x = x_155_cast_fp16)[name = tensor("linear_75_cast_fp16")]; + tensor x_157_cast_fp16 = add(x = x_151_cast_fp16, y = linear_75_cast_fp16)[name = tensor("x_157_cast_fp16")]; + tensor var_1460_axes_0 = const()[name = tensor("op_1460_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320557312)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320559424)))]; + tensor var_1460_cast_fp16 = layer_norm(axes = var_1460_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1386_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = tensor("op_1460_cast_fp16")]; + tensor var_1469_to_fp16 = const()[name = tensor("op_1469_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320561536)))]; + tensor var_1470_to_fp16 = const()[name = tensor("op_1470_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328950208)))]; + tensor linear_76_cast_fp16 = linear(bias = var_1470_to_fp16, weight = var_1469_to_fp16, x = var_1460_cast_fp16)[name = tensor("linear_76_cast_fp16")]; + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast_fp16 = gelu(mode = x_161_mode_0, x = linear_76_cast_fp16)[name = tensor("x_161_cast_fp16")]; + tensor var_1475_to_fp16 = const()[name = tensor("op_1475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328958464)))]; + tensor var_1476_to_fp16 = const()[name = tensor("op_1476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337347136)))]; + tensor linear_77_cast_fp16 = linear(bias = var_1476_to_fp16, weight = var_1475_to_fp16, x = x_161_cast_fp16)[name = tensor("linear_77_cast_fp16")]; + tensor x_163_cast_fp16 = add(x = x_157_cast_fp16, y = linear_77_cast_fp16)[name = tensor("x_163_cast_fp16")]; + tensor var_1486 = const()[name = tensor("op_1486"), val = tensor(-1)]; + tensor var_1502_axes_0 = const()[name = tensor("op_1502_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337349248)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337351360)))]; + tensor var_1492_to_fp16 = const()[name = tensor("op_1492_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1502_cast_fp16 = layer_norm(axes = var_1502_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1492_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = tensor("op_1502_cast_fp16")]; + tensor var_1513_to_fp16 = const()[name = tensor("op_1513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337353472)))]; + tensor var_1514_to_fp16 = const()[name = tensor("op_1514_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339450688)))]; + tensor linear_78_cast_fp16 = linear(bias = var_1514_to_fp16, weight = var_1513_to_fp16, x = var_1502_cast_fp16)[name = tensor("linear_78_cast_fp16")]; + tensor var_1517_to_fp16 = const()[name = tensor("op_1517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339452800)))]; + tensor linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1517_to_fp16, x = var_1502_cast_fp16)[name = tensor("linear_79_cast_fp16")]; + tensor var_1521_to_fp16 = const()[name = tensor("op_1521_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341550016)))]; + tensor var_1522_to_fp16 = const()[name = tensor("op_1522_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343647232)))]; + tensor linear_80_cast_fp16 = linear(bias = var_1522_to_fp16, weight = var_1521_to_fp16, x = var_1502_cast_fp16)[name = tensor("linear_80_cast_fp16")]; + tensor var_1530 = const()[name = tensor("op_1530"), val = tensor([1, 1500, 16, -1])]; + tensor var_1531_cast_fp16 = reshape(shape = var_1530, x = linear_78_cast_fp16)[name = tensor("op_1531_cast_fp16")]; + tensor const_194_to_fp16 = const()[name = tensor("const_194_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast_fp16 = mul(x = var_1531_cast_fp16, y = const_194_to_fp16)[name = tensor("q_55_cast_fp16")]; + tensor var_1537 = const()[name = tensor("op_1537"), val = tensor([1, 1500, 16, -1])]; + tensor var_1538_cast_fp16 = reshape(shape = var_1537, x = linear_79_cast_fp16)[name = tensor("op_1538_cast_fp16")]; + tensor const_195_to_fp16 = const()[name = tensor("const_195_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast_fp16 = mul(x = var_1538_cast_fp16, y = const_195_to_fp16)[name = tensor("k_55_cast_fp16")]; + tensor var_1544 = const()[name = tensor("op_1544"), val = tensor([1, 1500, 16, -1])]; + tensor var_1545_cast_fp16 = reshape(shape = var_1544, x = linear_80_cast_fp16)[name = tensor("op_1545_cast_fp16")]; + tensor var_1546 = const()[name = tensor("op_1546"), val = tensor([0, 2, 1, 3])]; + tensor qk_27_transpose_x_0 = const()[name = tensor("qk_27_transpose_x_0"), val = tensor(false)]; + tensor qk_27_transpose_y_0 = const()[name = tensor("qk_27_transpose_y_0"), val = tensor(false)]; + tensor transpose_122_perm_0 = const()[name = tensor("transpose_122_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_123_perm_0 = const()[name = tensor("transpose_123_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_185 = transpose(perm = transpose_123_perm_0, x = k_55_cast_fp16)[name = tensor("transpose_185")]; + tensor transpose_186 = transpose(perm = transpose_122_perm_0, x = q_55_cast_fp16)[name = tensor("transpose_186")]; + tensor qk_27_cast_fp16 = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_186, y = transpose_185)[name = tensor("qk_27_cast_fp16")]; + tensor var_1550_cast_fp16 = softmax(axis = var_1486, x = qk_27_cast_fp16)[name = tensor("op_1550_cast_fp16")]; + tensor var_1552_transpose_x_0 = const()[name = tensor("op_1552_transpose_x_0"), val = tensor(false)]; + tensor var_1552_transpose_y_0 = const()[name = tensor("op_1552_transpose_y_0"), val = tensor(false)]; + tensor transpose_187 = transpose(perm = var_1546, x = var_1545_cast_fp16)[name = tensor("transpose_187")]; + tensor var_1552_cast_fp16 = matmul(transpose_x = var_1552_transpose_x_0, transpose_y = var_1552_transpose_y_0, x = var_1550_cast_fp16, y = transpose_187)[name = tensor("op_1552_cast_fp16")]; + tensor var_1553 = const()[name = tensor("op_1553"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = tensor("concat_13"), val = tensor([1, 1500, 1024])]; + tensor transpose_184 = transpose(perm = var_1553, x = var_1552_cast_fp16)[name = tensor("transpose_184")]; + tensor x_167_cast_fp16 = reshape(shape = concat_13, x = transpose_184)[name = tensor("x_167_cast_fp16")]; + tensor var_1558_to_fp16 = const()[name = tensor("op_1558_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343649344)))]; + tensor var_1559_to_fp16 = const()[name = tensor("op_1559_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345746560)))]; + tensor linear_81_cast_fp16 = linear(bias = var_1559_to_fp16, weight = var_1558_to_fp16, x = x_167_cast_fp16)[name = tensor("linear_81_cast_fp16")]; + tensor x_169_cast_fp16 = add(x = x_163_cast_fp16, y = linear_81_cast_fp16)[name = tensor("x_169_cast_fp16")]; + tensor var_1566_axes_0 = const()[name = tensor("op_1566_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345748672)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345750784)))]; + tensor var_1566_cast_fp16 = layer_norm(axes = var_1566_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1492_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = tensor("op_1566_cast_fp16")]; + tensor var_1575_to_fp16 = const()[name = tensor("op_1575_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345752896)))]; + tensor var_1576_to_fp16 = const()[name = tensor("op_1576_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354141568)))]; + tensor linear_82_cast_fp16 = linear(bias = var_1576_to_fp16, weight = var_1575_to_fp16, x = var_1566_cast_fp16)[name = tensor("linear_82_cast_fp16")]; + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast_fp16 = gelu(mode = x_173_mode_0, x = linear_82_cast_fp16)[name = tensor("x_173_cast_fp16")]; + tensor var_1581_to_fp16 = const()[name = tensor("op_1581_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354149824)))]; + tensor var_1582_to_fp16 = const()[name = tensor("op_1582_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362538496)))]; + tensor linear_83_cast_fp16 = linear(bias = var_1582_to_fp16, weight = var_1581_to_fp16, x = x_173_cast_fp16)[name = tensor("linear_83_cast_fp16")]; + tensor x_175_cast_fp16 = add(x = x_169_cast_fp16, y = linear_83_cast_fp16)[name = tensor("x_175_cast_fp16")]; + tensor var_1592 = const()[name = tensor("op_1592"), val = tensor(-1)]; + tensor var_1608_axes_0 = const()[name = tensor("op_1608_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362540608)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362542720)))]; + tensor var_1598_to_fp16 = const()[name = tensor("op_1598_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1608_cast_fp16 = layer_norm(axes = var_1608_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1598_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = tensor("op_1608_cast_fp16")]; + tensor var_1619_to_fp16 = const()[name = tensor("op_1619_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362544832)))]; + tensor var_1620_to_fp16 = const()[name = tensor("op_1620_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364642048)))]; + tensor linear_84_cast_fp16 = linear(bias = var_1620_to_fp16, weight = var_1619_to_fp16, x = var_1608_cast_fp16)[name = tensor("linear_84_cast_fp16")]; + tensor var_1623_to_fp16 = const()[name = tensor("op_1623_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364644160)))]; + tensor linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1623_to_fp16, x = var_1608_cast_fp16)[name = tensor("linear_85_cast_fp16")]; + tensor var_1627_to_fp16 = const()[name = tensor("op_1627_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366741376)))]; + tensor var_1628_to_fp16 = const()[name = tensor("op_1628_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368838592)))]; + tensor linear_86_cast_fp16 = linear(bias = var_1628_to_fp16, weight = var_1627_to_fp16, x = var_1608_cast_fp16)[name = tensor("linear_86_cast_fp16")]; + tensor var_1636 = const()[name = tensor("op_1636"), val = tensor([1, 1500, 16, -1])]; + tensor var_1637_cast_fp16 = reshape(shape = var_1636, x = linear_84_cast_fp16)[name = tensor("op_1637_cast_fp16")]; + tensor const_196_to_fp16 = const()[name = tensor("const_196_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast_fp16 = mul(x = var_1637_cast_fp16, y = const_196_to_fp16)[name = tensor("q_59_cast_fp16")]; + tensor var_1643 = const()[name = tensor("op_1643"), val = tensor([1, 1500, 16, -1])]; + tensor var_1644_cast_fp16 = reshape(shape = var_1643, x = linear_85_cast_fp16)[name = tensor("op_1644_cast_fp16")]; + tensor const_197_to_fp16 = const()[name = tensor("const_197_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast_fp16 = mul(x = var_1644_cast_fp16, y = const_197_to_fp16)[name = tensor("k_59_cast_fp16")]; + tensor var_1650 = const()[name = tensor("op_1650"), val = tensor([1, 1500, 16, -1])]; + tensor var_1651_cast_fp16 = reshape(shape = var_1650, x = linear_86_cast_fp16)[name = tensor("op_1651_cast_fp16")]; + tensor var_1652 = const()[name = tensor("op_1652"), val = tensor([0, 2, 1, 3])]; + tensor qk_29_transpose_x_0 = const()[name = tensor("qk_29_transpose_x_0"), val = tensor(false)]; + tensor qk_29_transpose_y_0 = const()[name = tensor("qk_29_transpose_y_0"), val = tensor(false)]; + tensor transpose_124_perm_0 = const()[name = tensor("transpose_124_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_125_perm_0 = const()[name = tensor("transpose_125_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_181 = transpose(perm = transpose_125_perm_0, x = k_59_cast_fp16)[name = tensor("transpose_181")]; + tensor transpose_182 = transpose(perm = transpose_124_perm_0, x = q_59_cast_fp16)[name = tensor("transpose_182")]; + tensor qk_29_cast_fp16 = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_182, y = transpose_181)[name = tensor("qk_29_cast_fp16")]; + tensor var_1656_cast_fp16 = softmax(axis = var_1592, x = qk_29_cast_fp16)[name = tensor("op_1656_cast_fp16")]; + tensor var_1658_transpose_x_0 = const()[name = tensor("op_1658_transpose_x_0"), val = tensor(false)]; + tensor var_1658_transpose_y_0 = const()[name = tensor("op_1658_transpose_y_0"), val = tensor(false)]; + tensor transpose_183 = transpose(perm = var_1652, x = var_1651_cast_fp16)[name = tensor("transpose_183")]; + tensor var_1658_cast_fp16 = matmul(transpose_x = var_1658_transpose_x_0, transpose_y = var_1658_transpose_y_0, x = var_1656_cast_fp16, y = transpose_183)[name = tensor("op_1658_cast_fp16")]; + tensor var_1659 = const()[name = tensor("op_1659"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = tensor("concat_14"), val = tensor([1, 1500, 1024])]; + tensor transpose_180 = transpose(perm = var_1659, x = var_1658_cast_fp16)[name = tensor("transpose_180")]; + tensor x_179_cast_fp16 = reshape(shape = concat_14, x = transpose_180)[name = tensor("x_179_cast_fp16")]; + tensor var_1664_to_fp16 = const()[name = tensor("op_1664_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368840704)))]; + tensor var_1665_to_fp16 = const()[name = tensor("op_1665_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370937920)))]; + tensor linear_87_cast_fp16 = linear(bias = var_1665_to_fp16, weight = var_1664_to_fp16, x = x_179_cast_fp16)[name = tensor("linear_87_cast_fp16")]; + tensor x_181_cast_fp16 = add(x = x_175_cast_fp16, y = linear_87_cast_fp16)[name = tensor("x_181_cast_fp16")]; + tensor var_1672_axes_0 = const()[name = tensor("op_1672_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370940032)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370942144)))]; + tensor var_1672_cast_fp16 = layer_norm(axes = var_1672_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1598_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = tensor("op_1672_cast_fp16")]; + tensor var_1681_to_fp16 = const()[name = tensor("op_1681_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370944256)))]; + tensor var_1682_to_fp16 = const()[name = tensor("op_1682_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379332928)))]; + tensor linear_88_cast_fp16 = linear(bias = var_1682_to_fp16, weight = var_1681_to_fp16, x = var_1672_cast_fp16)[name = tensor("linear_88_cast_fp16")]; + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast_fp16 = gelu(mode = x_185_mode_0, x = linear_88_cast_fp16)[name = tensor("x_185_cast_fp16")]; + tensor var_1687_to_fp16 = const()[name = tensor("op_1687_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379341184)))]; + tensor var_1688_to_fp16 = const()[name = tensor("op_1688_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387729856)))]; + tensor linear_89_cast_fp16 = linear(bias = var_1688_to_fp16, weight = var_1687_to_fp16, x = x_185_cast_fp16)[name = tensor("linear_89_cast_fp16")]; + tensor x_187_cast_fp16 = add(x = x_181_cast_fp16, y = linear_89_cast_fp16)[name = tensor("x_187_cast_fp16")]; + tensor var_1698 = const()[name = tensor("op_1698"), val = tensor(-1)]; + tensor var_1714_axes_0 = const()[name = tensor("op_1714_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387731968)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387734080)))]; + tensor var_1704_to_fp16 = const()[name = tensor("op_1704_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1714_cast_fp16 = layer_norm(axes = var_1714_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1704_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = tensor("op_1714_cast_fp16")]; + tensor var_1725_to_fp16 = const()[name = tensor("op_1725_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387736192)))]; + tensor var_1726_to_fp16 = const()[name = tensor("op_1726_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389833408)))]; + tensor linear_90_cast_fp16 = linear(bias = var_1726_to_fp16, weight = var_1725_to_fp16, x = var_1714_cast_fp16)[name = tensor("linear_90_cast_fp16")]; + tensor var_1729_to_fp16 = const()[name = tensor("op_1729_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389835520)))]; + tensor linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1729_to_fp16, x = var_1714_cast_fp16)[name = tensor("linear_91_cast_fp16")]; + tensor var_1733_to_fp16 = const()[name = tensor("op_1733_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391932736)))]; + tensor var_1734_to_fp16 = const()[name = tensor("op_1734_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394029952)))]; + tensor linear_92_cast_fp16 = linear(bias = var_1734_to_fp16, weight = var_1733_to_fp16, x = var_1714_cast_fp16)[name = tensor("linear_92_cast_fp16")]; + tensor var_1742 = const()[name = tensor("op_1742"), val = tensor([1, 1500, 16, -1])]; + tensor var_1743_cast_fp16 = reshape(shape = var_1742, x = linear_90_cast_fp16)[name = tensor("op_1743_cast_fp16")]; + tensor const_198_to_fp16 = const()[name = tensor("const_198_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast_fp16 = mul(x = var_1743_cast_fp16, y = const_198_to_fp16)[name = tensor("q_63_cast_fp16")]; + tensor var_1749 = const()[name = tensor("op_1749"), val = tensor([1, 1500, 16, -1])]; + tensor var_1750_cast_fp16 = reshape(shape = var_1749, x = linear_91_cast_fp16)[name = tensor("op_1750_cast_fp16")]; + tensor const_199_to_fp16 = const()[name = tensor("const_199_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast_fp16 = mul(x = var_1750_cast_fp16, y = const_199_to_fp16)[name = tensor("k_63_cast_fp16")]; + tensor var_1756 = const()[name = tensor("op_1756"), val = tensor([1, 1500, 16, -1])]; + tensor var_1757_cast_fp16 = reshape(shape = var_1756, x = linear_92_cast_fp16)[name = tensor("op_1757_cast_fp16")]; + tensor var_1758 = const()[name = tensor("op_1758"), val = tensor([0, 2, 1, 3])]; + tensor qk_31_transpose_x_0 = const()[name = tensor("qk_31_transpose_x_0"), val = tensor(false)]; + tensor qk_31_transpose_y_0 = const()[name = tensor("qk_31_transpose_y_0"), val = tensor(false)]; + tensor transpose_126_perm_0 = const()[name = tensor("transpose_126_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_127_perm_0 = const()[name = tensor("transpose_127_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_177 = transpose(perm = transpose_127_perm_0, x = k_63_cast_fp16)[name = tensor("transpose_177")]; + tensor transpose_178 = transpose(perm = transpose_126_perm_0, x = q_63_cast_fp16)[name = tensor("transpose_178")]; + tensor qk_31_cast_fp16 = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_178, y = transpose_177)[name = tensor("qk_31_cast_fp16")]; + tensor var_1762_cast_fp16 = softmax(axis = var_1698, x = qk_31_cast_fp16)[name = tensor("op_1762_cast_fp16")]; + tensor var_1764_transpose_x_0 = const()[name = tensor("op_1764_transpose_x_0"), val = tensor(false)]; + tensor var_1764_transpose_y_0 = const()[name = tensor("op_1764_transpose_y_0"), val = tensor(false)]; + tensor transpose_179 = transpose(perm = var_1758, x = var_1757_cast_fp16)[name = tensor("transpose_179")]; + tensor var_1764_cast_fp16 = matmul(transpose_x = var_1764_transpose_x_0, transpose_y = var_1764_transpose_y_0, x = var_1762_cast_fp16, y = transpose_179)[name = tensor("op_1764_cast_fp16")]; + tensor var_1765 = const()[name = tensor("op_1765"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = tensor("concat_15"), val = tensor([1, 1500, 1024])]; + tensor transpose_176 = transpose(perm = var_1765, x = var_1764_cast_fp16)[name = tensor("transpose_176")]; + tensor x_191_cast_fp16 = reshape(shape = concat_15, x = transpose_176)[name = tensor("x_191_cast_fp16")]; + tensor var_1770_to_fp16 = const()[name = tensor("op_1770_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394032064)))]; + tensor var_1771_to_fp16 = const()[name = tensor("op_1771_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396129280)))]; + tensor linear_93_cast_fp16 = linear(bias = var_1771_to_fp16, weight = var_1770_to_fp16, x = x_191_cast_fp16)[name = tensor("linear_93_cast_fp16")]; + tensor x_193_cast_fp16 = add(x = x_187_cast_fp16, y = linear_93_cast_fp16)[name = tensor("x_193_cast_fp16")]; + tensor var_1778_axes_0 = const()[name = tensor("op_1778_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396131392)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396133504)))]; + tensor var_1778_cast_fp16 = layer_norm(axes = var_1778_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1704_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16)[name = tensor("op_1778_cast_fp16")]; + tensor var_1787_to_fp16 = const()[name = tensor("op_1787_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396135616)))]; + tensor var_1788_to_fp16 = const()[name = tensor("op_1788_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404524288)))]; + tensor linear_94_cast_fp16 = linear(bias = var_1788_to_fp16, weight = var_1787_to_fp16, x = var_1778_cast_fp16)[name = tensor("linear_94_cast_fp16")]; + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast_fp16 = gelu(mode = x_197_mode_0, x = linear_94_cast_fp16)[name = tensor("x_197_cast_fp16")]; + tensor var_1793_to_fp16 = const()[name = tensor("op_1793_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404532544)))]; + tensor var_1794_to_fp16 = const()[name = tensor("op_1794_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412921216)))]; + tensor linear_95_cast_fp16 = linear(bias = var_1794_to_fp16, weight = var_1793_to_fp16, x = x_197_cast_fp16)[name = tensor("linear_95_cast_fp16")]; + tensor x_199_cast_fp16 = add(x = x_193_cast_fp16, y = linear_95_cast_fp16)[name = tensor("x_199_cast_fp16")]; + tensor var_1804 = const()[name = tensor("op_1804"), val = tensor(-1)]; + tensor var_1820_axes_0 = const()[name = tensor("op_1820_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412923328)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412925440)))]; + tensor var_1810_to_fp16 = const()[name = tensor("op_1810_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1820_cast_fp16 = layer_norm(axes = var_1820_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1810_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = tensor("op_1820_cast_fp16")]; + tensor var_1831_to_fp16 = const()[name = tensor("op_1831_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412927552)))]; + tensor var_1832_to_fp16 = const()[name = tensor("op_1832_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415024768)))]; + tensor linear_96_cast_fp16 = linear(bias = var_1832_to_fp16, weight = var_1831_to_fp16, x = var_1820_cast_fp16)[name = tensor("linear_96_cast_fp16")]; + tensor var_1835_to_fp16 = const()[name = tensor("op_1835_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415026880)))]; + tensor linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1835_to_fp16, x = var_1820_cast_fp16)[name = tensor("linear_97_cast_fp16")]; + tensor var_1839_to_fp16 = const()[name = tensor("op_1839_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417124096)))]; + tensor var_1840_to_fp16 = const()[name = tensor("op_1840_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419221312)))]; + tensor linear_98_cast_fp16 = linear(bias = var_1840_to_fp16, weight = var_1839_to_fp16, x = var_1820_cast_fp16)[name = tensor("linear_98_cast_fp16")]; + tensor var_1848 = const()[name = tensor("op_1848"), val = tensor([1, 1500, 16, -1])]; + tensor var_1849_cast_fp16 = reshape(shape = var_1848, x = linear_96_cast_fp16)[name = tensor("op_1849_cast_fp16")]; + tensor const_200_to_fp16 = const()[name = tensor("const_200_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_67_cast_fp16 = mul(x = var_1849_cast_fp16, y = const_200_to_fp16)[name = tensor("q_67_cast_fp16")]; + tensor var_1855 = const()[name = tensor("op_1855"), val = tensor([1, 1500, 16, -1])]; + tensor var_1856_cast_fp16 = reshape(shape = var_1855, x = linear_97_cast_fp16)[name = tensor("op_1856_cast_fp16")]; + tensor const_201_to_fp16 = const()[name = tensor("const_201_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_67_cast_fp16 = mul(x = var_1856_cast_fp16, y = const_201_to_fp16)[name = tensor("k_67_cast_fp16")]; + tensor var_1862 = const()[name = tensor("op_1862"), val = tensor([1, 1500, 16, -1])]; + tensor var_1863_cast_fp16 = reshape(shape = var_1862, x = linear_98_cast_fp16)[name = tensor("op_1863_cast_fp16")]; + tensor var_1864 = const()[name = tensor("op_1864"), val = tensor([0, 2, 1, 3])]; + tensor qk_33_transpose_x_0 = const()[name = tensor("qk_33_transpose_x_0"), val = tensor(false)]; + tensor qk_33_transpose_y_0 = const()[name = tensor("qk_33_transpose_y_0"), val = tensor(false)]; + tensor transpose_128_perm_0 = const()[name = tensor("transpose_128_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_129_perm_0 = const()[name = tensor("transpose_129_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_173 = transpose(perm = transpose_129_perm_0, x = k_67_cast_fp16)[name = tensor("transpose_173")]; + tensor transpose_174 = transpose(perm = transpose_128_perm_0, x = q_67_cast_fp16)[name = tensor("transpose_174")]; + tensor qk_33_cast_fp16 = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_174, y = transpose_173)[name = tensor("qk_33_cast_fp16")]; + tensor var_1868_cast_fp16 = softmax(axis = var_1804, x = qk_33_cast_fp16)[name = tensor("op_1868_cast_fp16")]; + tensor var_1870_transpose_x_0 = const()[name = tensor("op_1870_transpose_x_0"), val = tensor(false)]; + tensor var_1870_transpose_y_0 = const()[name = tensor("op_1870_transpose_y_0"), val = tensor(false)]; + tensor transpose_175 = transpose(perm = var_1864, x = var_1863_cast_fp16)[name = tensor("transpose_175")]; + tensor var_1870_cast_fp16 = matmul(transpose_x = var_1870_transpose_x_0, transpose_y = var_1870_transpose_y_0, x = var_1868_cast_fp16, y = transpose_175)[name = tensor("op_1870_cast_fp16")]; + tensor var_1871 = const()[name = tensor("op_1871"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = tensor("concat_16"), val = tensor([1, 1500, 1024])]; + tensor transpose_172 = transpose(perm = var_1871, x = var_1870_cast_fp16)[name = tensor("transpose_172")]; + tensor x_203_cast_fp16 = reshape(shape = concat_16, x = transpose_172)[name = tensor("x_203_cast_fp16")]; + tensor var_1876_to_fp16 = const()[name = tensor("op_1876_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419223424)))]; + tensor var_1877_to_fp16 = const()[name = tensor("op_1877_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421320640)))]; + tensor linear_99_cast_fp16 = linear(bias = var_1877_to_fp16, weight = var_1876_to_fp16, x = x_203_cast_fp16)[name = tensor("linear_99_cast_fp16")]; + tensor x_205_cast_fp16 = add(x = x_199_cast_fp16, y = linear_99_cast_fp16)[name = tensor("x_205_cast_fp16")]; + tensor var_1884_axes_0 = const()[name = tensor("op_1884_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421322752)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421324864)))]; + tensor var_1884_cast_fp16 = layer_norm(axes = var_1884_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1810_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = tensor("op_1884_cast_fp16")]; + tensor var_1893_to_fp16 = const()[name = tensor("op_1893_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421326976)))]; + tensor var_1894_to_fp16 = const()[name = tensor("op_1894_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429715648)))]; + tensor linear_100_cast_fp16 = linear(bias = var_1894_to_fp16, weight = var_1893_to_fp16, x = var_1884_cast_fp16)[name = tensor("linear_100_cast_fp16")]; + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast_fp16 = gelu(mode = x_209_mode_0, x = linear_100_cast_fp16)[name = tensor("x_209_cast_fp16")]; + tensor var_1899_to_fp16 = const()[name = tensor("op_1899_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429723904)))]; + tensor var_1900_to_fp16 = const()[name = tensor("op_1900_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438112576)))]; + tensor linear_101_cast_fp16 = linear(bias = var_1900_to_fp16, weight = var_1899_to_fp16, x = x_209_cast_fp16)[name = tensor("linear_101_cast_fp16")]; + tensor x_211_cast_fp16 = add(x = x_205_cast_fp16, y = linear_101_cast_fp16)[name = tensor("x_211_cast_fp16")]; + tensor var_1910 = const()[name = tensor("op_1910"), val = tensor(-1)]; + tensor var_1926_axes_0 = const()[name = tensor("op_1926_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438114688)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438116800)))]; + tensor var_1916_to_fp16 = const()[name = tensor("op_1916_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1926_cast_fp16 = layer_norm(axes = var_1926_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1916_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = tensor("op_1926_cast_fp16")]; + tensor var_1937_to_fp16 = const()[name = tensor("op_1937_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438118912)))]; + tensor var_1938_to_fp16 = const()[name = tensor("op_1938_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440216128)))]; + tensor linear_102_cast_fp16 = linear(bias = var_1938_to_fp16, weight = var_1937_to_fp16, x = var_1926_cast_fp16)[name = tensor("linear_102_cast_fp16")]; + tensor var_1941_to_fp16 = const()[name = tensor("op_1941_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440218240)))]; + tensor linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1941_to_fp16, x = var_1926_cast_fp16)[name = tensor("linear_103_cast_fp16")]; + tensor var_1945_to_fp16 = const()[name = tensor("op_1945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442315456)))]; + tensor var_1946_to_fp16 = const()[name = tensor("op_1946_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444412672)))]; + tensor linear_104_cast_fp16 = linear(bias = var_1946_to_fp16, weight = var_1945_to_fp16, x = var_1926_cast_fp16)[name = tensor("linear_104_cast_fp16")]; + tensor var_1954 = const()[name = tensor("op_1954"), val = tensor([1, 1500, 16, -1])]; + tensor var_1955_cast_fp16 = reshape(shape = var_1954, x = linear_102_cast_fp16)[name = tensor("op_1955_cast_fp16")]; + tensor const_202_to_fp16 = const()[name = tensor("const_202_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_71_cast_fp16 = mul(x = var_1955_cast_fp16, y = const_202_to_fp16)[name = tensor("q_71_cast_fp16")]; + tensor var_1961 = const()[name = tensor("op_1961"), val = tensor([1, 1500, 16, -1])]; + tensor var_1962_cast_fp16 = reshape(shape = var_1961, x = linear_103_cast_fp16)[name = tensor("op_1962_cast_fp16")]; + tensor const_203_to_fp16 = const()[name = tensor("const_203_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_71_cast_fp16 = mul(x = var_1962_cast_fp16, y = const_203_to_fp16)[name = tensor("k_71_cast_fp16")]; + tensor var_1968 = const()[name = tensor("op_1968"), val = tensor([1, 1500, 16, -1])]; + tensor var_1969_cast_fp16 = reshape(shape = var_1968, x = linear_104_cast_fp16)[name = tensor("op_1969_cast_fp16")]; + tensor var_1970 = const()[name = tensor("op_1970"), val = tensor([0, 2, 1, 3])]; + tensor qk_35_transpose_x_0 = const()[name = tensor("qk_35_transpose_x_0"), val = tensor(false)]; + tensor qk_35_transpose_y_0 = const()[name = tensor("qk_35_transpose_y_0"), val = tensor(false)]; + tensor transpose_130_perm_0 = const()[name = tensor("transpose_130_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_131_perm_0 = const()[name = tensor("transpose_131_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_169 = transpose(perm = transpose_131_perm_0, x = k_71_cast_fp16)[name = tensor("transpose_169")]; + tensor transpose_170 = transpose(perm = transpose_130_perm_0, x = q_71_cast_fp16)[name = tensor("transpose_170")]; + tensor qk_35_cast_fp16 = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_170, y = transpose_169)[name = tensor("qk_35_cast_fp16")]; + tensor var_1974_cast_fp16 = softmax(axis = var_1910, x = qk_35_cast_fp16)[name = tensor("op_1974_cast_fp16")]; + tensor var_1976_transpose_x_0 = const()[name = tensor("op_1976_transpose_x_0"), val = tensor(false)]; + tensor var_1976_transpose_y_0 = const()[name = tensor("op_1976_transpose_y_0"), val = tensor(false)]; + tensor transpose_171 = transpose(perm = var_1970, x = var_1969_cast_fp16)[name = tensor("transpose_171")]; + tensor var_1976_cast_fp16 = matmul(transpose_x = var_1976_transpose_x_0, transpose_y = var_1976_transpose_y_0, x = var_1974_cast_fp16, y = transpose_171)[name = tensor("op_1976_cast_fp16")]; + tensor var_1977 = const()[name = tensor("op_1977"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = tensor("concat_17"), val = tensor([1, 1500, 1024])]; + tensor transpose_168 = transpose(perm = var_1977, x = var_1976_cast_fp16)[name = tensor("transpose_168")]; + tensor x_215_cast_fp16 = reshape(shape = concat_17, x = transpose_168)[name = tensor("x_215_cast_fp16")]; + tensor var_1982_to_fp16 = const()[name = tensor("op_1982_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444414784)))]; + tensor var_1983_to_fp16 = const()[name = tensor("op_1983_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446512000)))]; + tensor linear_105_cast_fp16 = linear(bias = var_1983_to_fp16, weight = var_1982_to_fp16, x = x_215_cast_fp16)[name = tensor("linear_105_cast_fp16")]; + tensor x_217_cast_fp16 = add(x = x_211_cast_fp16, y = linear_105_cast_fp16)[name = tensor("x_217_cast_fp16")]; + tensor var_1990_axes_0 = const()[name = tensor("op_1990_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446514112)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446516224)))]; + tensor var_1990_cast_fp16 = layer_norm(axes = var_1990_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1916_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = tensor("op_1990_cast_fp16")]; + tensor var_1999_to_fp16 = const()[name = tensor("op_1999_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446518336)))]; + tensor var_2000_to_fp16 = const()[name = tensor("op_2000_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454907008)))]; + tensor linear_106_cast_fp16 = linear(bias = var_2000_to_fp16, weight = var_1999_to_fp16, x = var_1990_cast_fp16)[name = tensor("linear_106_cast_fp16")]; + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast_fp16 = gelu(mode = x_221_mode_0, x = linear_106_cast_fp16)[name = tensor("x_221_cast_fp16")]; + tensor var_2005_to_fp16 = const()[name = tensor("op_2005_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454915264)))]; + tensor var_2006_to_fp16 = const()[name = tensor("op_2006_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463303936)))]; + tensor linear_107_cast_fp16 = linear(bias = var_2006_to_fp16, weight = var_2005_to_fp16, x = x_221_cast_fp16)[name = tensor("linear_107_cast_fp16")]; + tensor x_223_cast_fp16 = add(x = x_217_cast_fp16, y = linear_107_cast_fp16)[name = tensor("x_223_cast_fp16")]; + tensor var_2016 = const()[name = tensor("op_2016"), val = tensor(-1)]; + tensor var_2032_axes_0 = const()[name = tensor("op_2032_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463306048)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463308160)))]; + tensor var_2022_to_fp16 = const()[name = tensor("op_2022_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2032_cast_fp16 = layer_norm(axes = var_2032_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2022_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = tensor("op_2032_cast_fp16")]; + tensor var_2043_to_fp16 = const()[name = tensor("op_2043_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463310272)))]; + tensor var_2044_to_fp16 = const()[name = tensor("op_2044_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465407488)))]; + tensor linear_108_cast_fp16 = linear(bias = var_2044_to_fp16, weight = var_2043_to_fp16, x = var_2032_cast_fp16)[name = tensor("linear_108_cast_fp16")]; + tensor var_2047_to_fp16 = const()[name = tensor("op_2047_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465409600)))]; + tensor linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2047_to_fp16, x = var_2032_cast_fp16)[name = tensor("linear_109_cast_fp16")]; + tensor var_2051_to_fp16 = const()[name = tensor("op_2051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(467506816)))]; + tensor var_2052_to_fp16 = const()[name = tensor("op_2052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469604032)))]; + tensor linear_110_cast_fp16 = linear(bias = var_2052_to_fp16, weight = var_2051_to_fp16, x = var_2032_cast_fp16)[name = tensor("linear_110_cast_fp16")]; + tensor var_2060 = const()[name = tensor("op_2060"), val = tensor([1, 1500, 16, -1])]; + tensor var_2061_cast_fp16 = reshape(shape = var_2060, x = linear_108_cast_fp16)[name = tensor("op_2061_cast_fp16")]; + tensor const_204_to_fp16 = const()[name = tensor("const_204_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_75_cast_fp16 = mul(x = var_2061_cast_fp16, y = const_204_to_fp16)[name = tensor("q_75_cast_fp16")]; + tensor var_2067 = const()[name = tensor("op_2067"), val = tensor([1, 1500, 16, -1])]; + tensor var_2068_cast_fp16 = reshape(shape = var_2067, x = linear_109_cast_fp16)[name = tensor("op_2068_cast_fp16")]; + tensor const_205_to_fp16 = const()[name = tensor("const_205_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_75_cast_fp16 = mul(x = var_2068_cast_fp16, y = const_205_to_fp16)[name = tensor("k_75_cast_fp16")]; + tensor var_2074 = const()[name = tensor("op_2074"), val = tensor([1, 1500, 16, -1])]; + tensor var_2075_cast_fp16 = reshape(shape = var_2074, x = linear_110_cast_fp16)[name = tensor("op_2075_cast_fp16")]; + tensor var_2076 = const()[name = tensor("op_2076"), val = tensor([0, 2, 1, 3])]; + tensor qk_37_transpose_x_0 = const()[name = tensor("qk_37_transpose_x_0"), val = tensor(false)]; + tensor qk_37_transpose_y_0 = const()[name = tensor("qk_37_transpose_y_0"), val = tensor(false)]; + tensor transpose_132_perm_0 = const()[name = tensor("transpose_132_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_133_perm_0 = const()[name = tensor("transpose_133_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_165 = transpose(perm = transpose_133_perm_0, x = k_75_cast_fp16)[name = tensor("transpose_165")]; + tensor transpose_166 = transpose(perm = transpose_132_perm_0, x = q_75_cast_fp16)[name = tensor("transpose_166")]; + tensor qk_37_cast_fp16 = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_166, y = transpose_165)[name = tensor("qk_37_cast_fp16")]; + tensor var_2080_cast_fp16 = softmax(axis = var_2016, x = qk_37_cast_fp16)[name = tensor("op_2080_cast_fp16")]; + tensor var_2082_transpose_x_0 = const()[name = tensor("op_2082_transpose_x_0"), val = tensor(false)]; + tensor var_2082_transpose_y_0 = const()[name = tensor("op_2082_transpose_y_0"), val = tensor(false)]; + tensor transpose_167 = transpose(perm = var_2076, x = var_2075_cast_fp16)[name = tensor("transpose_167")]; + tensor var_2082_cast_fp16 = matmul(transpose_x = var_2082_transpose_x_0, transpose_y = var_2082_transpose_y_0, x = var_2080_cast_fp16, y = transpose_167)[name = tensor("op_2082_cast_fp16")]; + tensor var_2083 = const()[name = tensor("op_2083"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = tensor("concat_18"), val = tensor([1, 1500, 1024])]; + tensor transpose_164 = transpose(perm = var_2083, x = var_2082_cast_fp16)[name = tensor("transpose_164")]; + tensor x_227_cast_fp16 = reshape(shape = concat_18, x = transpose_164)[name = tensor("x_227_cast_fp16")]; + tensor var_2088_to_fp16 = const()[name = tensor("op_2088_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469606144)))]; + tensor var_2089_to_fp16 = const()[name = tensor("op_2089_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471703360)))]; + tensor linear_111_cast_fp16 = linear(bias = var_2089_to_fp16, weight = var_2088_to_fp16, x = x_227_cast_fp16)[name = tensor("linear_111_cast_fp16")]; + tensor x_229_cast_fp16 = add(x = x_223_cast_fp16, y = linear_111_cast_fp16)[name = tensor("x_229_cast_fp16")]; + tensor var_2096_axes_0 = const()[name = tensor("op_2096_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471705472)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471707584)))]; + tensor var_2096_cast_fp16 = layer_norm(axes = var_2096_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2022_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = tensor("op_2096_cast_fp16")]; + tensor var_2105_to_fp16 = const()[name = tensor("op_2105_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471709696)))]; + tensor var_2106_to_fp16 = const()[name = tensor("op_2106_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480098368)))]; + tensor linear_112_cast_fp16 = linear(bias = var_2106_to_fp16, weight = var_2105_to_fp16, x = var_2096_cast_fp16)[name = tensor("linear_112_cast_fp16")]; + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast_fp16 = gelu(mode = x_233_mode_0, x = linear_112_cast_fp16)[name = tensor("x_233_cast_fp16")]; + tensor var_2111_to_fp16 = const()[name = tensor("op_2111_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480106624)))]; + tensor var_2112_to_fp16 = const()[name = tensor("op_2112_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488495296)))]; + tensor linear_113_cast_fp16 = linear(bias = var_2112_to_fp16, weight = var_2111_to_fp16, x = x_233_cast_fp16)[name = tensor("linear_113_cast_fp16")]; + tensor x_235_cast_fp16 = add(x = x_229_cast_fp16, y = linear_113_cast_fp16)[name = tensor("x_235_cast_fp16")]; + tensor var_2122 = const()[name = tensor("op_2122"), val = tensor(-1)]; + tensor var_2138_axes_0 = const()[name = tensor("op_2138_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488497408)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488499520)))]; + tensor var_2128_to_fp16 = const()[name = tensor("op_2128_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2138_cast_fp16 = layer_norm(axes = var_2138_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2128_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = tensor("op_2138_cast_fp16")]; + tensor var_2149_to_fp16 = const()[name = tensor("op_2149_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488501632)))]; + tensor var_2150_to_fp16 = const()[name = tensor("op_2150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490598848)))]; + tensor linear_114_cast_fp16 = linear(bias = var_2150_to_fp16, weight = var_2149_to_fp16, x = var_2138_cast_fp16)[name = tensor("linear_114_cast_fp16")]; + tensor var_2153_to_fp16 = const()[name = tensor("op_2153_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490600960)))]; + tensor linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2153_to_fp16, x = var_2138_cast_fp16)[name = tensor("linear_115_cast_fp16")]; + tensor var_2157_to_fp16 = const()[name = tensor("op_2157_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492698176)))]; + tensor var_2158_to_fp16 = const()[name = tensor("op_2158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494795392)))]; + tensor linear_116_cast_fp16 = linear(bias = var_2158_to_fp16, weight = var_2157_to_fp16, x = var_2138_cast_fp16)[name = tensor("linear_116_cast_fp16")]; + tensor var_2166 = const()[name = tensor("op_2166"), val = tensor([1, 1500, 16, -1])]; + tensor var_2167_cast_fp16 = reshape(shape = var_2166, x = linear_114_cast_fp16)[name = tensor("op_2167_cast_fp16")]; + tensor const_206_to_fp16 = const()[name = tensor("const_206_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_79_cast_fp16 = mul(x = var_2167_cast_fp16, y = const_206_to_fp16)[name = tensor("q_79_cast_fp16")]; + tensor var_2173 = const()[name = tensor("op_2173"), val = tensor([1, 1500, 16, -1])]; + tensor var_2174_cast_fp16 = reshape(shape = var_2173, x = linear_115_cast_fp16)[name = tensor("op_2174_cast_fp16")]; + tensor const_207_to_fp16 = const()[name = tensor("const_207_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_79_cast_fp16 = mul(x = var_2174_cast_fp16, y = const_207_to_fp16)[name = tensor("k_79_cast_fp16")]; + tensor var_2180 = const()[name = tensor("op_2180"), val = tensor([1, 1500, 16, -1])]; + tensor var_2181_cast_fp16 = reshape(shape = var_2180, x = linear_116_cast_fp16)[name = tensor("op_2181_cast_fp16")]; + tensor var_2182 = const()[name = tensor("op_2182"), val = tensor([0, 2, 1, 3])]; + tensor qk_39_transpose_x_0 = const()[name = tensor("qk_39_transpose_x_0"), val = tensor(false)]; + tensor qk_39_transpose_y_0 = const()[name = tensor("qk_39_transpose_y_0"), val = tensor(false)]; + tensor transpose_134_perm_0 = const()[name = tensor("transpose_134_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_135_perm_0 = const()[name = tensor("transpose_135_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_161 = transpose(perm = transpose_135_perm_0, x = k_79_cast_fp16)[name = tensor("transpose_161")]; + tensor transpose_162 = transpose(perm = transpose_134_perm_0, x = q_79_cast_fp16)[name = tensor("transpose_162")]; + tensor qk_39_cast_fp16 = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_162, y = transpose_161)[name = tensor("qk_39_cast_fp16")]; + tensor var_2186_cast_fp16 = softmax(axis = var_2122, x = qk_39_cast_fp16)[name = tensor("op_2186_cast_fp16")]; + tensor var_2188_transpose_x_0 = const()[name = tensor("op_2188_transpose_x_0"), val = tensor(false)]; + tensor var_2188_transpose_y_0 = const()[name = tensor("op_2188_transpose_y_0"), val = tensor(false)]; + tensor transpose_163 = transpose(perm = var_2182, x = var_2181_cast_fp16)[name = tensor("transpose_163")]; + tensor var_2188_cast_fp16 = matmul(transpose_x = var_2188_transpose_x_0, transpose_y = var_2188_transpose_y_0, x = var_2186_cast_fp16, y = transpose_163)[name = tensor("op_2188_cast_fp16")]; + tensor var_2189 = const()[name = tensor("op_2189"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = tensor("concat_19"), val = tensor([1, 1500, 1024])]; + tensor transpose_160 = transpose(perm = var_2189, x = var_2188_cast_fp16)[name = tensor("transpose_160")]; + tensor x_239_cast_fp16 = reshape(shape = concat_19, x = transpose_160)[name = tensor("x_239_cast_fp16")]; + tensor var_2194_to_fp16 = const()[name = tensor("op_2194_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494797504)))]; + tensor var_2195_to_fp16 = const()[name = tensor("op_2195_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496894720)))]; + tensor linear_117_cast_fp16 = linear(bias = var_2195_to_fp16, weight = var_2194_to_fp16, x = x_239_cast_fp16)[name = tensor("linear_117_cast_fp16")]; + tensor x_241_cast_fp16 = add(x = x_235_cast_fp16, y = linear_117_cast_fp16)[name = tensor("x_241_cast_fp16")]; + tensor var_2202_axes_0 = const()[name = tensor("op_2202_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496896832)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496898944)))]; + tensor var_2202_cast_fp16 = layer_norm(axes = var_2202_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2128_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = tensor("op_2202_cast_fp16")]; + tensor var_2211_to_fp16 = const()[name = tensor("op_2211_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496901056)))]; + tensor var_2212_to_fp16 = const()[name = tensor("op_2212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(505289728)))]; + tensor linear_118_cast_fp16 = linear(bias = var_2212_to_fp16, weight = var_2211_to_fp16, x = var_2202_cast_fp16)[name = tensor("linear_118_cast_fp16")]; + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast_fp16 = gelu(mode = x_245_mode_0, x = linear_118_cast_fp16)[name = tensor("x_245_cast_fp16")]; + tensor var_2217_to_fp16 = const()[name = tensor("op_2217_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(505297984)))]; + tensor var_2218_to_fp16 = const()[name = tensor("op_2218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513686656)))]; + tensor linear_119_cast_fp16 = linear(bias = var_2218_to_fp16, weight = var_2217_to_fp16, x = x_245_cast_fp16)[name = tensor("linear_119_cast_fp16")]; + tensor x_247_cast_fp16 = add(x = x_241_cast_fp16, y = linear_119_cast_fp16)[name = tensor("x_247_cast_fp16")]; + tensor var_2228 = const()[name = tensor("op_2228"), val = tensor(-1)]; + tensor var_2244_axes_0 = const()[name = tensor("op_2244_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513688768)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513690880)))]; + tensor var_2234_to_fp16 = const()[name = tensor("op_2234_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2244_cast_fp16 = layer_norm(axes = var_2244_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2234_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = tensor("op_2244_cast_fp16")]; + tensor var_2255_to_fp16 = const()[name = tensor("op_2255_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513692992)))]; + tensor var_2256_to_fp16 = const()[name = tensor("op_2256_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(515790208)))]; + tensor linear_120_cast_fp16 = linear(bias = var_2256_to_fp16, weight = var_2255_to_fp16, x = var_2244_cast_fp16)[name = tensor("linear_120_cast_fp16")]; + tensor var_2259_to_fp16 = const()[name = tensor("op_2259_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(515792320)))]; + tensor linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2259_to_fp16, x = var_2244_cast_fp16)[name = tensor("linear_121_cast_fp16")]; + tensor var_2263_to_fp16 = const()[name = tensor("op_2263_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517889536)))]; + tensor var_2264_to_fp16 = const()[name = tensor("op_2264_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(519986752)))]; + tensor linear_122_cast_fp16 = linear(bias = var_2264_to_fp16, weight = var_2263_to_fp16, x = var_2244_cast_fp16)[name = tensor("linear_122_cast_fp16")]; + tensor var_2272 = const()[name = tensor("op_2272"), val = tensor([1, 1500, 16, -1])]; + tensor var_2273_cast_fp16 = reshape(shape = var_2272, x = linear_120_cast_fp16)[name = tensor("op_2273_cast_fp16")]; + tensor const_208_to_fp16 = const()[name = tensor("const_208_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_83_cast_fp16 = mul(x = var_2273_cast_fp16, y = const_208_to_fp16)[name = tensor("q_83_cast_fp16")]; + tensor var_2279 = const()[name = tensor("op_2279"), val = tensor([1, 1500, 16, -1])]; + tensor var_2280_cast_fp16 = reshape(shape = var_2279, x = linear_121_cast_fp16)[name = tensor("op_2280_cast_fp16")]; + tensor const_209_to_fp16 = const()[name = tensor("const_209_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_83_cast_fp16 = mul(x = var_2280_cast_fp16, y = const_209_to_fp16)[name = tensor("k_83_cast_fp16")]; + tensor var_2286 = const()[name = tensor("op_2286"), val = tensor([1, 1500, 16, -1])]; + tensor var_2287_cast_fp16 = reshape(shape = var_2286, x = linear_122_cast_fp16)[name = tensor("op_2287_cast_fp16")]; + tensor var_2288 = const()[name = tensor("op_2288"), val = tensor([0, 2, 1, 3])]; + tensor qk_41_transpose_x_0 = const()[name = tensor("qk_41_transpose_x_0"), val = tensor(false)]; + tensor qk_41_transpose_y_0 = const()[name = tensor("qk_41_transpose_y_0"), val = tensor(false)]; + tensor transpose_136_perm_0 = const()[name = tensor("transpose_136_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_137_perm_0 = const()[name = tensor("transpose_137_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_157 = transpose(perm = transpose_137_perm_0, x = k_83_cast_fp16)[name = tensor("transpose_157")]; + tensor transpose_158 = transpose(perm = transpose_136_perm_0, x = q_83_cast_fp16)[name = tensor("transpose_158")]; + tensor qk_41_cast_fp16 = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_158, y = transpose_157)[name = tensor("qk_41_cast_fp16")]; + tensor var_2292_cast_fp16 = softmax(axis = var_2228, x = qk_41_cast_fp16)[name = tensor("op_2292_cast_fp16")]; + tensor var_2294_transpose_x_0 = const()[name = tensor("op_2294_transpose_x_0"), val = tensor(false)]; + tensor var_2294_transpose_y_0 = const()[name = tensor("op_2294_transpose_y_0"), val = tensor(false)]; + tensor transpose_159 = transpose(perm = var_2288, x = var_2287_cast_fp16)[name = tensor("transpose_159")]; + tensor var_2294_cast_fp16 = matmul(transpose_x = var_2294_transpose_x_0, transpose_y = var_2294_transpose_y_0, x = var_2292_cast_fp16, y = transpose_159)[name = tensor("op_2294_cast_fp16")]; + tensor var_2295 = const()[name = tensor("op_2295"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = tensor("concat_20"), val = tensor([1, 1500, 1024])]; + tensor transpose_156 = transpose(perm = var_2295, x = var_2294_cast_fp16)[name = tensor("transpose_156")]; + tensor x_251_cast_fp16 = reshape(shape = concat_20, x = transpose_156)[name = tensor("x_251_cast_fp16")]; + tensor var_2300_to_fp16 = const()[name = tensor("op_2300_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(519988864)))]; + tensor var_2301_to_fp16 = const()[name = tensor("op_2301_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522086080)))]; + tensor linear_123_cast_fp16 = linear(bias = var_2301_to_fp16, weight = var_2300_to_fp16, x = x_251_cast_fp16)[name = tensor("linear_123_cast_fp16")]; + tensor x_253_cast_fp16 = add(x = x_247_cast_fp16, y = linear_123_cast_fp16)[name = tensor("x_253_cast_fp16")]; + tensor var_2308_axes_0 = const()[name = tensor("op_2308_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522088192)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522090304)))]; + tensor var_2308_cast_fp16 = layer_norm(axes = var_2308_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2234_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = tensor("op_2308_cast_fp16")]; + tensor var_2317_to_fp16 = const()[name = tensor("op_2317_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522092416)))]; + tensor var_2318_to_fp16 = const()[name = tensor("op_2318_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530481088)))]; + tensor linear_124_cast_fp16 = linear(bias = var_2318_to_fp16, weight = var_2317_to_fp16, x = var_2308_cast_fp16)[name = tensor("linear_124_cast_fp16")]; + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast_fp16 = gelu(mode = x_257_mode_0, x = linear_124_cast_fp16)[name = tensor("x_257_cast_fp16")]; + tensor var_2323_to_fp16 = const()[name = tensor("op_2323_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530489344)))]; + tensor var_2324_to_fp16 = const()[name = tensor("op_2324_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538878016)))]; + tensor linear_125_cast_fp16 = linear(bias = var_2324_to_fp16, weight = var_2323_to_fp16, x = x_257_cast_fp16)[name = tensor("linear_125_cast_fp16")]; + tensor x_259_cast_fp16 = add(x = x_253_cast_fp16, y = linear_125_cast_fp16)[name = tensor("x_259_cast_fp16")]; + tensor var_2334 = const()[name = tensor("op_2334"), val = tensor(-1)]; + tensor var_2350_axes_0 = const()[name = tensor("op_2350_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538880128)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538882240)))]; + tensor var_2340_to_fp16 = const()[name = tensor("op_2340_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2350_cast_fp16 = layer_norm(axes = var_2350_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2340_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = tensor("op_2350_cast_fp16")]; + tensor var_2361_to_fp16 = const()[name = tensor("op_2361_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538884352)))]; + tensor var_2362_to_fp16 = const()[name = tensor("op_2362_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(540981568)))]; + tensor linear_126_cast_fp16 = linear(bias = var_2362_to_fp16, weight = var_2361_to_fp16, x = var_2350_cast_fp16)[name = tensor("linear_126_cast_fp16")]; + tensor var_2365_to_fp16 = const()[name = tensor("op_2365_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(540983680)))]; + tensor linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2365_to_fp16, x = var_2350_cast_fp16)[name = tensor("linear_127_cast_fp16")]; + tensor var_2369_to_fp16 = const()[name = tensor("op_2369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543080896)))]; + tensor var_2370_to_fp16 = const()[name = tensor("op_2370_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545178112)))]; + tensor linear_128_cast_fp16 = linear(bias = var_2370_to_fp16, weight = var_2369_to_fp16, x = var_2350_cast_fp16)[name = tensor("linear_128_cast_fp16")]; + tensor var_2378 = const()[name = tensor("op_2378"), val = tensor([1, 1500, 16, -1])]; + tensor var_2379_cast_fp16 = reshape(shape = var_2378, x = linear_126_cast_fp16)[name = tensor("op_2379_cast_fp16")]; + tensor const_210_to_fp16 = const()[name = tensor("const_210_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_87_cast_fp16 = mul(x = var_2379_cast_fp16, y = const_210_to_fp16)[name = tensor("q_87_cast_fp16")]; + tensor var_2385 = const()[name = tensor("op_2385"), val = tensor([1, 1500, 16, -1])]; + tensor var_2386_cast_fp16 = reshape(shape = var_2385, x = linear_127_cast_fp16)[name = tensor("op_2386_cast_fp16")]; + tensor const_211_to_fp16 = const()[name = tensor("const_211_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_87_cast_fp16 = mul(x = var_2386_cast_fp16, y = const_211_to_fp16)[name = tensor("k_87_cast_fp16")]; + tensor var_2392 = const()[name = tensor("op_2392"), val = tensor([1, 1500, 16, -1])]; + tensor var_2393_cast_fp16 = reshape(shape = var_2392, x = linear_128_cast_fp16)[name = tensor("op_2393_cast_fp16")]; + tensor var_2394 = const()[name = tensor("op_2394"), val = tensor([0, 2, 1, 3])]; + tensor qk_43_transpose_x_0 = const()[name = tensor("qk_43_transpose_x_0"), val = tensor(false)]; + tensor qk_43_transpose_y_0 = const()[name = tensor("qk_43_transpose_y_0"), val = tensor(false)]; + tensor transpose_138_perm_0 = const()[name = tensor("transpose_138_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_139_perm_0 = const()[name = tensor("transpose_139_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_153 = transpose(perm = transpose_139_perm_0, x = k_87_cast_fp16)[name = tensor("transpose_153")]; + tensor transpose_154 = transpose(perm = transpose_138_perm_0, x = q_87_cast_fp16)[name = tensor("transpose_154")]; + tensor qk_43_cast_fp16 = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_154, y = transpose_153)[name = tensor("qk_43_cast_fp16")]; + tensor var_2398_cast_fp16 = softmax(axis = var_2334, x = qk_43_cast_fp16)[name = tensor("op_2398_cast_fp16")]; + tensor var_2400_transpose_x_0 = const()[name = tensor("op_2400_transpose_x_0"), val = tensor(false)]; + tensor var_2400_transpose_y_0 = const()[name = tensor("op_2400_transpose_y_0"), val = tensor(false)]; + tensor transpose_155 = transpose(perm = var_2394, x = var_2393_cast_fp16)[name = tensor("transpose_155")]; + tensor var_2400_cast_fp16 = matmul(transpose_x = var_2400_transpose_x_0, transpose_y = var_2400_transpose_y_0, x = var_2398_cast_fp16, y = transpose_155)[name = tensor("op_2400_cast_fp16")]; + tensor var_2401 = const()[name = tensor("op_2401"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = tensor("concat_21"), val = tensor([1, 1500, 1024])]; + tensor transpose_152 = transpose(perm = var_2401, x = var_2400_cast_fp16)[name = tensor("transpose_152")]; + tensor x_263_cast_fp16 = reshape(shape = concat_21, x = transpose_152)[name = tensor("x_263_cast_fp16")]; + tensor var_2406_to_fp16 = const()[name = tensor("op_2406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545180224)))]; + tensor var_2407_to_fp16 = const()[name = tensor("op_2407_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547277440)))]; + tensor linear_129_cast_fp16 = linear(bias = var_2407_to_fp16, weight = var_2406_to_fp16, x = x_263_cast_fp16)[name = tensor("linear_129_cast_fp16")]; + tensor x_265_cast_fp16 = add(x = x_259_cast_fp16, y = linear_129_cast_fp16)[name = tensor("x_265_cast_fp16")]; + tensor var_2414_axes_0 = const()[name = tensor("op_2414_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547279552)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547281664)))]; + tensor var_2414_cast_fp16 = layer_norm(axes = var_2414_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2340_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = tensor("op_2414_cast_fp16")]; + tensor var_2423_to_fp16 = const()[name = tensor("op_2423_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547283776)))]; + tensor var_2424_to_fp16 = const()[name = tensor("op_2424_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555672448)))]; + tensor linear_130_cast_fp16 = linear(bias = var_2424_to_fp16, weight = var_2423_to_fp16, x = var_2414_cast_fp16)[name = tensor("linear_130_cast_fp16")]; + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast_fp16 = gelu(mode = x_269_mode_0, x = linear_130_cast_fp16)[name = tensor("x_269_cast_fp16")]; + tensor var_2429_to_fp16 = const()[name = tensor("op_2429_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555680704)))]; + tensor var_2430_to_fp16 = const()[name = tensor("op_2430_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564069376)))]; + tensor linear_131_cast_fp16 = linear(bias = var_2430_to_fp16, weight = var_2429_to_fp16, x = x_269_cast_fp16)[name = tensor("linear_131_cast_fp16")]; + tensor x_271_cast_fp16 = add(x = x_265_cast_fp16, y = linear_131_cast_fp16)[name = tensor("x_271_cast_fp16")]; + tensor var_2440 = const()[name = tensor("op_2440"), val = tensor(-1)]; + tensor var_2456_axes_0 = const()[name = tensor("op_2456_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564071488)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564073600)))]; + tensor var_2446_to_fp16 = const()[name = tensor("op_2446_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2456_cast_fp16 = layer_norm(axes = var_2456_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2446_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = tensor("op_2456_cast_fp16")]; + tensor var_2467_to_fp16 = const()[name = tensor("op_2467_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564075712)))]; + tensor var_2468_to_fp16 = const()[name = tensor("op_2468_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566172928)))]; + tensor linear_132_cast_fp16 = linear(bias = var_2468_to_fp16, weight = var_2467_to_fp16, x = var_2456_cast_fp16)[name = tensor("linear_132_cast_fp16")]; + tensor var_2471_to_fp16 = const()[name = tensor("op_2471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566175040)))]; + tensor linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2471_to_fp16, x = var_2456_cast_fp16)[name = tensor("linear_133_cast_fp16")]; + tensor var_2475_to_fp16 = const()[name = tensor("op_2475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568272256)))]; + tensor var_2476_to_fp16 = const()[name = tensor("op_2476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570369472)))]; + tensor linear_134_cast_fp16 = linear(bias = var_2476_to_fp16, weight = var_2475_to_fp16, x = var_2456_cast_fp16)[name = tensor("linear_134_cast_fp16")]; + tensor var_2484 = const()[name = tensor("op_2484"), val = tensor([1, 1500, 16, -1])]; + tensor var_2485_cast_fp16 = reshape(shape = var_2484, x = linear_132_cast_fp16)[name = tensor("op_2485_cast_fp16")]; + tensor const_212_to_fp16 = const()[name = tensor("const_212_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_91_cast_fp16 = mul(x = var_2485_cast_fp16, y = const_212_to_fp16)[name = tensor("q_91_cast_fp16")]; + tensor var_2491 = const()[name = tensor("op_2491"), val = tensor([1, 1500, 16, -1])]; + tensor var_2492_cast_fp16 = reshape(shape = var_2491, x = linear_133_cast_fp16)[name = tensor("op_2492_cast_fp16")]; + tensor const_213_to_fp16 = const()[name = tensor("const_213_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_91_cast_fp16 = mul(x = var_2492_cast_fp16, y = const_213_to_fp16)[name = tensor("k_91_cast_fp16")]; + tensor var_2498 = const()[name = tensor("op_2498"), val = tensor([1, 1500, 16, -1])]; + tensor var_2499_cast_fp16 = reshape(shape = var_2498, x = linear_134_cast_fp16)[name = tensor("op_2499_cast_fp16")]; + tensor var_2500 = const()[name = tensor("op_2500"), val = tensor([0, 2, 1, 3])]; + tensor qk_45_transpose_x_0 = const()[name = tensor("qk_45_transpose_x_0"), val = tensor(false)]; + tensor qk_45_transpose_y_0 = const()[name = tensor("qk_45_transpose_y_0"), val = tensor(false)]; + tensor transpose_140_perm_0 = const()[name = tensor("transpose_140_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_141_perm_0 = const()[name = tensor("transpose_141_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_149 = transpose(perm = transpose_141_perm_0, x = k_91_cast_fp16)[name = tensor("transpose_149")]; + tensor transpose_150 = transpose(perm = transpose_140_perm_0, x = q_91_cast_fp16)[name = tensor("transpose_150")]; + tensor qk_45_cast_fp16 = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_150, y = transpose_149)[name = tensor("qk_45_cast_fp16")]; + tensor var_2504_cast_fp16 = softmax(axis = var_2440, x = qk_45_cast_fp16)[name = tensor("op_2504_cast_fp16")]; + tensor var_2506_transpose_x_0 = const()[name = tensor("op_2506_transpose_x_0"), val = tensor(false)]; + tensor var_2506_transpose_y_0 = const()[name = tensor("op_2506_transpose_y_0"), val = tensor(false)]; + tensor transpose_151 = transpose(perm = var_2500, x = var_2499_cast_fp16)[name = tensor("transpose_151")]; + tensor var_2506_cast_fp16 = matmul(transpose_x = var_2506_transpose_x_0, transpose_y = var_2506_transpose_y_0, x = var_2504_cast_fp16, y = transpose_151)[name = tensor("op_2506_cast_fp16")]; + tensor var_2507 = const()[name = tensor("op_2507"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = tensor("concat_22"), val = tensor([1, 1500, 1024])]; + tensor transpose_148 = transpose(perm = var_2507, x = var_2506_cast_fp16)[name = tensor("transpose_148")]; + tensor x_275_cast_fp16 = reshape(shape = concat_22, x = transpose_148)[name = tensor("x_275_cast_fp16")]; + tensor var_2512_to_fp16 = const()[name = tensor("op_2512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570371584)))]; + tensor var_2513_to_fp16 = const()[name = tensor("op_2513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572468800)))]; + tensor linear_135_cast_fp16 = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = x_275_cast_fp16)[name = tensor("linear_135_cast_fp16")]; + tensor x_277_cast_fp16 = add(x = x_271_cast_fp16, y = linear_135_cast_fp16)[name = tensor("x_277_cast_fp16")]; + tensor var_2520_axes_0 = const()[name = tensor("op_2520_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572470912)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572473024)))]; + tensor var_2520_cast_fp16 = layer_norm(axes = var_2520_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2446_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = tensor("op_2520_cast_fp16")]; + tensor var_2529_to_fp16 = const()[name = tensor("op_2529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572475136)))]; + tensor var_2530_to_fp16 = const()[name = tensor("op_2530_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580863808)))]; + tensor linear_136_cast_fp16 = linear(bias = var_2530_to_fp16, weight = var_2529_to_fp16, x = var_2520_cast_fp16)[name = tensor("linear_136_cast_fp16")]; + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast_fp16 = gelu(mode = x_281_mode_0, x = linear_136_cast_fp16)[name = tensor("x_281_cast_fp16")]; + tensor var_2535_to_fp16 = const()[name = tensor("op_2535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580872064)))]; + tensor var_2536_to_fp16 = const()[name = tensor("op_2536_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589260736)))]; + tensor linear_137_cast_fp16 = linear(bias = var_2536_to_fp16, weight = var_2535_to_fp16, x = x_281_cast_fp16)[name = tensor("linear_137_cast_fp16")]; + tensor x_283_cast_fp16 = add(x = x_277_cast_fp16, y = linear_137_cast_fp16)[name = tensor("x_283_cast_fp16")]; + tensor var_2546 = const()[name = tensor("op_2546"), val = tensor(-1)]; + tensor var_2562_axes_0 = const()[name = tensor("op_2562_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589262848)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589264960)))]; + tensor var_2552_to_fp16 = const()[name = tensor("op_2552_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2562_cast_fp16 = layer_norm(axes = var_2562_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2552_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = tensor("op_2562_cast_fp16")]; + tensor var_2573_to_fp16 = const()[name = tensor("op_2573_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589267072)))]; + tensor var_2574_to_fp16 = const()[name = tensor("op_2574_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591364288)))]; + tensor linear_138_cast_fp16 = linear(bias = var_2574_to_fp16, weight = var_2573_to_fp16, x = var_2562_cast_fp16)[name = tensor("linear_138_cast_fp16")]; + tensor var_2577_to_fp16 = const()[name = tensor("op_2577_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591366400)))]; + tensor linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2577_to_fp16, x = var_2562_cast_fp16)[name = tensor("linear_139_cast_fp16")]; + tensor var_2581_to_fp16 = const()[name = tensor("op_2581_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593463616)))]; + tensor var_2582_to_fp16 = const()[name = tensor("op_2582_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595560832)))]; + tensor linear_140_cast_fp16 = linear(bias = var_2582_to_fp16, weight = var_2581_to_fp16, x = var_2562_cast_fp16)[name = tensor("linear_140_cast_fp16")]; + tensor var_2590 = const()[name = tensor("op_2590"), val = tensor([1, 1500, 16, -1])]; + tensor var_2591_cast_fp16 = reshape(shape = var_2590, x = linear_138_cast_fp16)[name = tensor("op_2591_cast_fp16")]; + tensor const_214_to_fp16 = const()[name = tensor("const_214_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast_fp16 = mul(x = var_2591_cast_fp16, y = const_214_to_fp16)[name = tensor("q_cast_fp16")]; + tensor var_2597 = const()[name = tensor("op_2597"), val = tensor([1, 1500, 16, -1])]; + tensor var_2598_cast_fp16 = reshape(shape = var_2597, x = linear_139_cast_fp16)[name = tensor("op_2598_cast_fp16")]; + tensor const_215_to_fp16 = const()[name = tensor("const_215_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast_fp16 = mul(x = var_2598_cast_fp16, y = const_215_to_fp16)[name = tensor("k_cast_fp16")]; + tensor var_2604 = const()[name = tensor("op_2604"), val = tensor([1, 1500, 16, -1])]; + tensor var_2605_cast_fp16 = reshape(shape = var_2604, x = linear_140_cast_fp16)[name = tensor("op_2605_cast_fp16")]; + tensor var_2606 = const()[name = tensor("op_2606"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_142_perm_0 = const()[name = tensor("transpose_142_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_143_perm_0 = const()[name = tensor("transpose_143_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_145 = transpose(perm = transpose_143_perm_0, x = k_cast_fp16)[name = tensor("transpose_145")]; + tensor transpose_146 = transpose(perm = transpose_142_perm_0, x = q_cast_fp16)[name = tensor("transpose_146")]; + tensor qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_146, y = transpose_145)[name = tensor("qk_cast_fp16")]; + tensor var_2610_cast_fp16 = softmax(axis = var_2546, x = qk_cast_fp16)[name = tensor("op_2610_cast_fp16")]; + tensor var_2612_transpose_x_0 = const()[name = tensor("op_2612_transpose_x_0"), val = tensor(false)]; + tensor var_2612_transpose_y_0 = const()[name = tensor("op_2612_transpose_y_0"), val = tensor(false)]; + tensor transpose_147 = transpose(perm = var_2606, x = var_2605_cast_fp16)[name = tensor("transpose_147")]; + tensor var_2612_cast_fp16 = matmul(transpose_x = var_2612_transpose_x_0, transpose_y = var_2612_transpose_y_0, x = var_2610_cast_fp16, y = transpose_147)[name = tensor("op_2612_cast_fp16")]; + tensor var_2613 = const()[name = tensor("op_2613"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = tensor("concat_23"), val = tensor([1, 1500, 1024])]; + tensor transpose_144 = transpose(perm = var_2613, x = var_2612_cast_fp16)[name = tensor("transpose_144")]; + tensor x_287_cast_fp16 = reshape(shape = concat_23, x = transpose_144)[name = tensor("x_287_cast_fp16")]; + tensor var_2618_to_fp16 = const()[name = tensor("op_2618_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595562944)))]; + tensor var_2619_to_fp16 = const()[name = tensor("op_2619_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597660160)))]; + tensor linear_141_cast_fp16 = linear(bias = var_2619_to_fp16, weight = var_2618_to_fp16, x = x_287_cast_fp16)[name = tensor("linear_141_cast_fp16")]; + tensor x_289_cast_fp16 = add(x = x_283_cast_fp16, y = linear_141_cast_fp16)[name = tensor("x_289_cast_fp16")]; + tensor var_2626_axes_0 = const()[name = tensor("op_2626_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597662272)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597664384)))]; + tensor var_2626_cast_fp16 = layer_norm(axes = var_2626_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2552_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = tensor("op_2626_cast_fp16")]; + tensor var_2635_to_fp16 = const()[name = tensor("op_2635_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597666496)))]; + tensor var_2636_to_fp16 = const()[name = tensor("op_2636_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606055168)))]; + tensor linear_142_cast_fp16 = linear(bias = var_2636_to_fp16, weight = var_2635_to_fp16, x = var_2626_cast_fp16)[name = tensor("linear_142_cast_fp16")]; + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast_fp16 = gelu(mode = x_293_mode_0, x = linear_142_cast_fp16)[name = tensor("x_293_cast_fp16")]; + tensor var_2641_to_fp16 = const()[name = tensor("op_2641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606063424)))]; + tensor var_2642_to_fp16 = const()[name = tensor("op_2642_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614452096)))]; + tensor linear_143_cast_fp16 = linear(bias = var_2642_to_fp16, weight = var_2641_to_fp16, x = x_293_cast_fp16)[name = tensor("linear_143_cast_fp16")]; + tensor x_cast_fp16 = add(x = x_289_cast_fp16, y = linear_143_cast_fp16)[name = tensor("x_cast_fp16")]; + tensor var_2655_axes_0 = const()[name = tensor("op_2655_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614454208)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614456320)))]; + tensor var_2646_to_fp16 = const()[name = tensor("op_2646_to_fp16"), val = tensor(0x1.5p-17)]; + tensor output = layer_norm(axes = var_2655_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_2646_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = tensor("op_2655_cast_fp16")]; + } -> (output); +} \ No newline at end of file