diff --git "a/ggml-medium.en-encoder.mlmodelc/model.mil" "b/ggml-medium.en-encoder.mlmodelc/model.mil"
new file mode 100644--- /dev/null
+++ "b/ggml-medium.en-encoder.mlmodelc/model.mil"
@@ -0,0 +1,1455 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})]
+{
+    func main<ios15>(tensor<fp32, [1, 80, 3000]> logmel_data) {
+            tensor<int32, []> var_56 = const()[name = tensor<string, []>("op_56"), val = tensor<int32, []>(1)];
+            tensor<int32, [1]> var_64 = const()[name = tensor<string, []>("op_64"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_66 = const()[name = tensor<string, []>("op_66"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> var_68_pad_type_0 = const()[name = tensor<string, []>("op_68_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> var_68_pad_0 = const()[name = tensor<string, []>("op_68_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<string, []> logmel_data_to_fp16_dtype_0 = const()[name = tensor<string, []>("logmel_data_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [1024, 80, 3]> weight_3_to_fp16 = const()[name = tensor<string, []>("weight_3_to_fp16"), val = tensor<fp16, [1024, 80, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [1024]> bias_3_to_fp16 = const()[name = tensor<string, []>("bias_3_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491648)))];
+            tensor<fp16, [1, 80, 3000]> cast_727 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data);
+            tensor<fp16, [1, 1024, 3000]> var_68_cast = 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 = cast_727);
+            tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1024, 3000]> input_1_cast = gelu(mode = input_1_mode_0, x = var_68_cast);
+            tensor<int32, []> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<int32, []>(1)];
+            tensor<int32, [1]> var_81 = const()[name = tensor<string, []>("op_81"), val = tensor<int32, [1]>([2])];
+            tensor<int32, [1]> var_83 = const()[name = tensor<string, []>("op_83"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> var_85_pad_type_0 = const()[name = tensor<string, []>("op_85_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> var_85_pad_0 = const()[name = tensor<string, []>("op_85_pad_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [1024, 1024, 3]> weight_7_to_fp16 = const()[name = tensor<string, []>("weight_7_to_fp16"), val = tensor<fp16, [1024, 1024, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493760)))];
+            tensor<fp16, [1024]> bias_7_to_fp16 = const()[name = tensor<string, []>("bias_7_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6785280)))];
+            tensor<fp16, [1, 1024, 1500]> var_85_cast = conv(bias = bias_7_to_fp16, dilations = var_83, groups = var_72, pad = var_85_pad_0, pad_type = var_85_pad_type_0, strides = var_81, weight = weight_7_to_fp16, x = input_1_cast);
+            tensor<string, []> x_3_mode_0 = const()[name = tensor<string, []>("x_3_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1024, 1500]> x_3_cast = gelu(mode = x_3_mode_0, x = var_85_cast);
+            tensor<int32, [3]> var_90 = const()[name = tensor<string, []>("op_90"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1500, 1024]> positional_embedding_to_fp16 = const()[name = tensor<string, []>("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6787392)))];
+            tensor<fp16, [1, 1500, 1024]> transpose_192 = transpose(perm = var_90, x = x_3_cast);
+            tensor<fp16, [1, 1500, 1024]> var_93_cast = add(x = transpose_192, y = positional_embedding_to_fp16);
+            tensor<int32, []> var_106 = const()[name = tensor<string, []>("op_106"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_123_axes_0 = const()[name = tensor<string, []>("op_123_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9859456)))];
+            tensor<fp16, [1024]> blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9861568)))];
+            tensor<fp16, []> var_112_to_fp16 = const()[name = tensor<string, []>("op_112_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_123_cast = layer_norm(axes = var_123_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_112_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_93_cast);
+            tensor<fp16, [1024, 1024]> var_134_to_fp16 = const()[name = tensor<string, []>("op_134_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9863680)))];
+            tensor<fp16, [1024]> var_135_to_fp16 = const()[name = tensor<string, []>("op_135_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11960896)))];
+            tensor<fp16, [1, 1500, 1024]> q_1_cast = linear(bias = var_135_to_fp16, weight = var_134_to_fp16, x = var_123_cast);
+            tensor<fp16, [1024, 1024]> var_138_to_fp16 = const()[name = tensor<string, []>("op_138_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11963008)))];
+            tensor<fp16, [1024]> k_1_bias_0_to_fp16 = const()[name = tensor<string, []>("k_1_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14060224)))];
+            tensor<fp16, [1, 1500, 1024]> k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_138_to_fp16, x = var_123_cast);
+            tensor<fp16, [1024, 1024]> var_142_to_fp16 = const()[name = tensor<string, []>("op_142_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14062336)))];
+            tensor<fp16, [1024]> var_143_to_fp16 = const()[name = tensor<string, []>("op_143_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16159552)))];
+            tensor<fp16, [1, 1500, 1024]> v_1_cast = linear(bias = var_143_to_fp16, weight = var_142_to_fp16, x = var_123_cast);
+            tensor<int32, [4]> var_151 = const()[name = tensor<string, []>("op_151"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_152_cast = reshape(shape = var_151, x = q_1_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_168_to_fp16 = const()[name = tensor<string, []>("const_168_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_3_cast = mul(x = var_152_cast, y = const_168_to_fp16);
+            tensor<int32, [4]> var_158 = const()[name = tensor<string, []>("op_158"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_159_cast = reshape(shape = var_158, x = k_1_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_169_to_fp16 = const()[name = tensor<string, []>("const_169_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_3_cast = mul(x = var_159_cast, y = const_169_to_fp16);
+            tensor<int32, [4]> var_165 = const()[name = tensor<string, []>("op_165"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_166_cast = reshape(shape = var_165, x = v_1_cast);
+            tensor<int32, [4]> var_167 = const()[name = tensor<string, []>("op_167"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_1_transpose_x_0 = const()[name = tensor<string, []>("qk_1_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_1_transpose_y_0 = const()[name = tensor<string, []>("qk_1_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_48_perm_0 = const()[name = tensor<string, []>("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_49_perm_0 = const()[name = tensor<string, []>("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_189 = transpose(perm = transpose_49_perm_0, x = k_3_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_190 = transpose(perm = transpose_48_perm_0, x = q_3_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_190, y = transpose_189);
+            tensor<fp16, [1, 16, 1500, 1500]> var_171_cast = softmax(axis = var_106, x = qk_1_cast);
+            tensor<bool, []> var_173_transpose_x_0 = const()[name = tensor<string, []>("op_173_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_173_transpose_y_0 = const()[name = tensor<string, []>("op_173_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_191 = transpose(perm = var_167, x = var_166_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_173_cast = matmul(transpose_x = var_173_transpose_x_0, transpose_y = var_173_transpose_y_0, x = var_171_cast, y = transpose_191);
+            tensor<int32, [4]> var_174 = const()[name = tensor<string, []>("op_174"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_0 = const()[name = tensor<string, []>("concat_0"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_188 = transpose(perm = var_174, x = var_173_cast);
+            tensor<fp16, [1, 1500, 1024]> x_11_cast = reshape(shape = concat_0, x = transpose_188);
+            tensor<fp16, [1024, 1024]> var_179_to_fp16 = const()[name = tensor<string, []>("op_179_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16161664)))];
+            tensor<fp16, [1024]> var_180_to_fp16 = const()[name = tensor<string, []>("op_180_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18258880)))];
+            tensor<fp16, [1, 1500, 1024]> var_181_cast = linear(bias = var_180_to_fp16, weight = var_179_to_fp16, x = x_11_cast);
+            tensor<fp16, [1, 1500, 1024]> x_13_cast = add(x = var_93_cast, y = var_181_cast);
+            tensor<int32, [1]> var_187_axes_0 = const()[name = tensor<string, []>("op_187_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18260992)))];
+            tensor<fp16, [1024]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18263104)))];
+            tensor<fp16, [1, 1500, 1024]> var_187_cast = layer_norm(axes = var_187_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_112_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast);
+            tensor<fp16, [4096, 1024]> var_196_to_fp16 = const()[name = tensor<string, []>("op_196_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18265216)))];
+            tensor<fp16, [4096]> var_197_to_fp16 = const()[name = tensor<string, []>("op_197_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26653888)))];
+            tensor<fp16, [1, 1500, 4096]> input_9_cast = linear(bias = var_197_to_fp16, weight = var_196_to_fp16, x = var_187_cast);
+            tensor<string, []> x_17_mode_0 = const()[name = tensor<string, []>("x_17_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast);
+            tensor<fp16, [1024, 4096]> var_202_to_fp16 = const()[name = tensor<string, []>("op_202_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26662144)))];
+            tensor<fp16, [1024]> var_203_to_fp16 = const()[name = tensor<string, []>("op_203_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35050816)))];
+            tensor<fp16, [1, 1500, 1024]> var_204_cast = linear(bias = var_203_to_fp16, weight = var_202_to_fp16, x = x_17_cast);
+            tensor<fp16, [1, 1500, 1024]> x_19_cast = add(x = x_13_cast, y = var_204_cast);
+            tensor<int32, []> var_213 = const()[name = tensor<string, []>("op_213"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_230_axes_0 = const()[name = tensor<string, []>("op_230_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35052928)))];
+            tensor<fp16, [1024]> blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35055040)))];
+            tensor<fp16, []> var_219_to_fp16 = const()[name = tensor<string, []>("op_219_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_230_cast = layer_norm(axes = var_230_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_219_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast);
+            tensor<fp16, [1024, 1024]> var_241_to_fp16 = const()[name = tensor<string, []>("op_241_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35057152)))];
+            tensor<fp16, [1024]> var_242_to_fp16 = const()[name = tensor<string, []>("op_242_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37154368)))];
+            tensor<fp16, [1, 1500, 1024]> q_5_cast = linear(bias = var_242_to_fp16, weight = var_241_to_fp16, x = var_230_cast);
+            tensor<fp16, [1024, 1024]> var_245_to_fp16 = const()[name = tensor<string, []>("op_245_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37156480)))];
+            tensor<fp16, [1024]> k_5_bias_0_to_fp16 = const()[name = tensor<string, []>("k_5_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39253696)))];
+            tensor<fp16, [1, 1500, 1024]> k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_245_to_fp16, x = var_230_cast);
+            tensor<fp16, [1024, 1024]> var_249_to_fp16 = const()[name = tensor<string, []>("op_249_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39255808)))];
+            tensor<fp16, [1024]> var_250_to_fp16 = const()[name = tensor<string, []>("op_250_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41353024)))];
+            tensor<fp16, [1, 1500, 1024]> v_5_cast = linear(bias = var_250_to_fp16, weight = var_249_to_fp16, x = var_230_cast);
+            tensor<int32, [4]> var_258 = const()[name = tensor<string, []>("op_258"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_259_cast = reshape(shape = var_258, x = q_5_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_170_to_fp16 = const()[name = tensor<string, []>("const_170_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_7_cast = mul(x = var_259_cast, y = const_170_to_fp16);
+            tensor<int32, [4]> var_265 = const()[name = tensor<string, []>("op_265"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_266_cast = reshape(shape = var_265, x = k_5_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_171_to_fp16 = const()[name = tensor<string, []>("const_171_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_7_cast = mul(x = var_266_cast, y = const_171_to_fp16);
+            tensor<int32, [4]> var_272 = const()[name = tensor<string, []>("op_272"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_273_cast = reshape(shape = var_272, x = v_5_cast);
+            tensor<int32, [4]> var_274 = const()[name = tensor<string, []>("op_274"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_3_transpose_x_0 = const()[name = tensor<string, []>("qk_3_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_3_transpose_y_0 = const()[name = tensor<string, []>("qk_3_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_50_perm_0 = const()[name = tensor<string, []>("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_51_perm_0 = const()[name = tensor<string, []>("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_185 = transpose(perm = transpose_51_perm_0, x = k_7_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_186 = transpose(perm = transpose_50_perm_0, x = q_7_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_186, y = transpose_185);
+            tensor<fp16, [1, 16, 1500, 1500]> var_278_cast = softmax(axis = var_213, x = qk_3_cast);
+            tensor<bool, []> var_280_transpose_x_0 = const()[name = tensor<string, []>("op_280_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_280_transpose_y_0 = const()[name = tensor<string, []>("op_280_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_187 = transpose(perm = var_274, x = var_273_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_280_cast = matmul(transpose_x = var_280_transpose_x_0, transpose_y = var_280_transpose_y_0, x = var_278_cast, y = transpose_187);
+            tensor<int32, [4]> var_281 = const()[name = tensor<string, []>("op_281"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_1 = const()[name = tensor<string, []>("concat_1"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_184 = transpose(perm = var_281, x = var_280_cast);
+            tensor<fp16, [1, 1500, 1024]> x_23_cast = reshape(shape = concat_1, x = transpose_184);
+            tensor<fp16, [1024, 1024]> var_286_to_fp16 = const()[name = tensor<string, []>("op_286_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41355136)))];
+            tensor<fp16, [1024]> var_287_to_fp16 = const()[name = tensor<string, []>("op_287_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43452352)))];
+            tensor<fp16, [1, 1500, 1024]> var_288_cast = linear(bias = var_287_to_fp16, weight = var_286_to_fp16, x = x_23_cast);
+            tensor<fp16, [1, 1500, 1024]> x_25_cast = add(x = x_19_cast, y = var_288_cast);
+            tensor<int32, [1]> var_294_axes_0 = const()[name = tensor<string, []>("op_294_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43454464)))];
+            tensor<fp16, [1024]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43456576)))];
+            tensor<fp16, [1, 1500, 1024]> var_294_cast = layer_norm(axes = var_294_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_219_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast);
+            tensor<fp16, [4096, 1024]> var_303_to_fp16 = const()[name = tensor<string, []>("op_303_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43458688)))];
+            tensor<fp16, [4096]> var_304_to_fp16 = const()[name = tensor<string, []>("op_304_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51847360)))];
+            tensor<fp16, [1, 1500, 4096]> input_17_cast = linear(bias = var_304_to_fp16, weight = var_303_to_fp16, x = var_294_cast);
+            tensor<string, []> x_29_mode_0 = const()[name = tensor<string, []>("x_29_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast);
+            tensor<fp16, [1024, 4096]> var_309_to_fp16 = const()[name = tensor<string, []>("op_309_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51855616)))];
+            tensor<fp16, [1024]> var_310_to_fp16 = const()[name = tensor<string, []>("op_310_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60244288)))];
+            tensor<fp16, [1, 1500, 1024]> var_311_cast = linear(bias = var_310_to_fp16, weight = var_309_to_fp16, x = x_29_cast);
+            tensor<fp16, [1, 1500, 1024]> x_31_cast = add(x = x_25_cast, y = var_311_cast);
+            tensor<int32, []> var_320 = const()[name = tensor<string, []>("op_320"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_337_axes_0 = const()[name = tensor<string, []>("op_337_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60246400)))];
+            tensor<fp16, [1024]> blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60248512)))];
+            tensor<fp16, []> var_326_to_fp16 = const()[name = tensor<string, []>("op_326_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_337_cast = layer_norm(axes = var_337_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);
+            tensor<fp16, [1024, 1024]> var_348_to_fp16 = const()[name = tensor<string, []>("op_348_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60250624)))];
+            tensor<fp16, [1024]> var_349_to_fp16 = const()[name = tensor<string, []>("op_349_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62347840)))];
+            tensor<fp16, [1, 1500, 1024]> q_9_cast = linear(bias = var_349_to_fp16, weight = var_348_to_fp16, x = var_337_cast);
+            tensor<fp16, [1024, 1024]> var_352_to_fp16 = const()[name = tensor<string, []>("op_352_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62349952)))];
+            tensor<fp16, [1024]> k_9_bias_0_to_fp16 = const()[name = tensor<string, []>("k_9_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64447168)))];
+            tensor<fp16, [1, 1500, 1024]> k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_352_to_fp16, x = var_337_cast);
+            tensor<fp16, [1024, 1024]> var_356_to_fp16 = const()[name = tensor<string, []>("op_356_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64449280)))];
+            tensor<fp16, [1024]> var_357_to_fp16 = const()[name = tensor<string, []>("op_357_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66546496)))];
+            tensor<fp16, [1, 1500, 1024]> v_9_cast = linear(bias = var_357_to_fp16, weight = var_356_to_fp16, x = var_337_cast);
+            tensor<int32, [4]> var_365 = const()[name = tensor<string, []>("op_365"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_366_cast = reshape(shape = var_365, x = q_9_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_172_to_fp16 = const()[name = tensor<string, []>("const_172_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_11_cast = mul(x = var_366_cast, y = const_172_to_fp16);
+            tensor<int32, [4]> var_372 = const()[name = tensor<string, []>("op_372"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_373_cast = reshape(shape = var_372, x = k_9_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_173_to_fp16 = const()[name = tensor<string, []>("const_173_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_11_cast = mul(x = var_373_cast, y = const_173_to_fp16);
+            tensor<int32, [4]> var_379 = const()[name = tensor<string, []>("op_379"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_380_cast = reshape(shape = var_379, x = v_9_cast);
+            tensor<int32, [4]> var_381 = const()[name = tensor<string, []>("op_381"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_5_transpose_x_0 = const()[name = tensor<string, []>("qk_5_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_5_transpose_y_0 = const()[name = tensor<string, []>("qk_5_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_52_perm_0 = const()[name = tensor<string, []>("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_53_perm_0 = const()[name = tensor<string, []>("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_181 = transpose(perm = transpose_53_perm_0, x = k_11_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_182 = transpose(perm = transpose_52_perm_0, x = q_11_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_182, y = transpose_181);
+            tensor<fp16, [1, 16, 1500, 1500]> var_385_cast = softmax(axis = var_320, x = qk_5_cast);
+            tensor<bool, []> var_387_transpose_x_0 = const()[name = tensor<string, []>("op_387_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_387_transpose_y_0 = const()[name = tensor<string, []>("op_387_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_183 = transpose(perm = var_381, x = var_380_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_387_cast = matmul(transpose_x = var_387_transpose_x_0, transpose_y = var_387_transpose_y_0, x = var_385_cast, y = transpose_183);
+            tensor<int32, [4]> var_388 = const()[name = tensor<string, []>("op_388"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_2 = const()[name = tensor<string, []>("concat_2"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_180 = transpose(perm = var_388, x = var_387_cast);
+            tensor<fp16, [1, 1500, 1024]> x_35_cast = reshape(shape = concat_2, x = transpose_180);
+            tensor<fp16, [1024, 1024]> var_393_to_fp16 = const()[name = tensor<string, []>("op_393_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66548608)))];
+            tensor<fp16, [1024]> var_394_to_fp16 = const()[name = tensor<string, []>("op_394_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68645824)))];
+            tensor<fp16, [1, 1500, 1024]> var_395_cast = linear(bias = var_394_to_fp16, weight = var_393_to_fp16, x = x_35_cast);
+            tensor<fp16, [1, 1500, 1024]> x_37_cast = add(x = x_31_cast, y = var_395_cast);
+            tensor<int32, [1]> var_401_axes_0 = const()[name = tensor<string, []>("op_401_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68647936)))];
+            tensor<fp16, [1024]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68650048)))];
+            tensor<fp16, [1, 1500, 1024]> var_401_cast = layer_norm(axes = var_401_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);
+            tensor<fp16, [4096, 1024]> var_410_to_fp16 = const()[name = tensor<string, []>("op_410_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68652160)))];
+            tensor<fp16, [4096]> var_411_to_fp16 = const()[name = tensor<string, []>("op_411_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77040832)))];
+            tensor<fp16, [1, 1500, 4096]> input_25_cast = linear(bias = var_411_to_fp16, weight = var_410_to_fp16, x = var_401_cast);
+            tensor<string, []> x_41_mode_0 = const()[name = tensor<string, []>("x_41_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast);
+            tensor<fp16, [1024, 4096]> var_416_to_fp16 = const()[name = tensor<string, []>("op_416_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77049088)))];
+            tensor<fp16, [1024]> var_417_to_fp16 = const()[name = tensor<string, []>("op_417_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85437760)))];
+            tensor<fp16, [1, 1500, 1024]> var_418_cast = linear(bias = var_417_to_fp16, weight = var_416_to_fp16, x = x_41_cast);
+            tensor<fp16, [1, 1500, 1024]> x_43_cast = add(x = x_37_cast, y = var_418_cast);
+            tensor<int32, []> var_427 = const()[name = tensor<string, []>("op_427"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_444_axes_0 = const()[name = tensor<string, []>("op_444_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85439872)))];
+            tensor<fp16, [1024]> blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85441984)))];
+            tensor<fp16, []> var_433_to_fp16 = const()[name = tensor<string, []>("op_433_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_444_cast = layer_norm(axes = var_444_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_433_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast);
+            tensor<fp16, [1024, 1024]> var_455_to_fp16 = const()[name = tensor<string, []>("op_455_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85444096)))];
+            tensor<fp16, [1024]> var_456_to_fp16 = const()[name = tensor<string, []>("op_456_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87541312)))];
+            tensor<fp16, [1, 1500, 1024]> q_13_cast = linear(bias = var_456_to_fp16, weight = var_455_to_fp16, x = var_444_cast);
+            tensor<fp16, [1024, 1024]> var_459_to_fp16 = const()[name = tensor<string, []>("op_459_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87543424)))];
+            tensor<fp16, [1024]> k_13_bias_0_to_fp16 = const()[name = tensor<string, []>("k_13_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89640640)))];
+            tensor<fp16, [1, 1500, 1024]> k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_459_to_fp16, x = var_444_cast);
+            tensor<fp16, [1024, 1024]> var_463_to_fp16 = const()[name = tensor<string, []>("op_463_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89642752)))];
+            tensor<fp16, [1024]> var_464_to_fp16 = const()[name = tensor<string, []>("op_464_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91739968)))];
+            tensor<fp16, [1, 1500, 1024]> v_13_cast = linear(bias = var_464_to_fp16, weight = var_463_to_fp16, x = var_444_cast);
+            tensor<int32, [4]> var_472 = const()[name = tensor<string, []>("op_472"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_473_cast = reshape(shape = var_472, x = q_13_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_174_to_fp16 = const()[name = tensor<string, []>("const_174_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_15_cast = mul(x = var_473_cast, y = const_174_to_fp16);
+            tensor<int32, [4]> var_479 = const()[name = tensor<string, []>("op_479"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_480_cast = reshape(shape = var_479, x = k_13_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_175_to_fp16 = const()[name = tensor<string, []>("const_175_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_15_cast = mul(x = var_480_cast, y = const_175_to_fp16);
+            tensor<int32, [4]> var_486 = const()[name = tensor<string, []>("op_486"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_487_cast = reshape(shape = var_486, x = v_13_cast);
+            tensor<int32, [4]> var_488 = const()[name = tensor<string, []>("op_488"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_7_transpose_x_0 = const()[name = tensor<string, []>("qk_7_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_7_transpose_y_0 = const()[name = tensor<string, []>("qk_7_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_54_perm_0 = const()[name = tensor<string, []>("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_55_perm_0 = const()[name = tensor<string, []>("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_177 = transpose(perm = transpose_55_perm_0, x = k_15_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_178 = transpose(perm = transpose_54_perm_0, x = q_15_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_178, y = transpose_177);
+            tensor<fp16, [1, 16, 1500, 1500]> var_492_cast = softmax(axis = var_427, x = qk_7_cast);
+            tensor<bool, []> var_494_transpose_x_0 = const()[name = tensor<string, []>("op_494_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_494_transpose_y_0 = const()[name = tensor<string, []>("op_494_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_179 = transpose(perm = var_488, x = var_487_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_494_cast = matmul(transpose_x = var_494_transpose_x_0, transpose_y = var_494_transpose_y_0, x = var_492_cast, y = transpose_179);
+            tensor<int32, [4]> var_495 = const()[name = tensor<string, []>("op_495"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_3 = const()[name = tensor<string, []>("concat_3"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_176 = transpose(perm = var_495, x = var_494_cast);
+            tensor<fp16, [1, 1500, 1024]> x_47_cast = reshape(shape = concat_3, x = transpose_176);
+            tensor<fp16, [1024, 1024]> var_500_to_fp16 = const()[name = tensor<string, []>("op_500_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91742080)))];
+            tensor<fp16, [1024]> var_501_to_fp16 = const()[name = tensor<string, []>("op_501_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93839296)))];
+            tensor<fp16, [1, 1500, 1024]> var_502_cast = linear(bias = var_501_to_fp16, weight = var_500_to_fp16, x = x_47_cast);
+            tensor<fp16, [1, 1500, 1024]> x_49_cast = add(x = x_43_cast, y = var_502_cast);
+            tensor<int32, [1]> var_508_axes_0 = const()[name = tensor<string, []>("op_508_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93841408)))];
+            tensor<fp16, [1024]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93843520)))];
+            tensor<fp16, [1, 1500, 1024]> var_508_cast = layer_norm(axes = var_508_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_433_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast);
+            tensor<fp16, [4096, 1024]> var_517_to_fp16 = const()[name = tensor<string, []>("op_517_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93845632)))];
+            tensor<fp16, [4096]> var_518_to_fp16 = const()[name = tensor<string, []>("op_518_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102234304)))];
+            tensor<fp16, [1, 1500, 4096]> input_33_cast = linear(bias = var_518_to_fp16, weight = var_517_to_fp16, x = var_508_cast);
+            tensor<string, []> x_53_mode_0 = const()[name = tensor<string, []>("x_53_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast);
+            tensor<fp16, [1024, 4096]> var_523_to_fp16 = const()[name = tensor<string, []>("op_523_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102242560)))];
+            tensor<fp16, [1024]> var_524_to_fp16 = const()[name = tensor<string, []>("op_524_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110631232)))];
+            tensor<fp16, [1, 1500, 1024]> var_525_cast = linear(bias = var_524_to_fp16, weight = var_523_to_fp16, x = x_53_cast);
+            tensor<fp16, [1, 1500, 1024]> x_55_cast = add(x = x_49_cast, y = var_525_cast);
+            tensor<int32, []> var_534 = const()[name = tensor<string, []>("op_534"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_551_axes_0 = const()[name = tensor<string, []>("op_551_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110633344)))];
+            tensor<fp16, [1024]> blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110635456)))];
+            tensor<fp16, []> var_540_to_fp16 = const()[name = tensor<string, []>("op_540_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_551_cast = layer_norm(axes = var_551_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_540_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast);
+            tensor<fp16, [1024, 1024]> var_562_to_fp16 = const()[name = tensor<string, []>("op_562_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110637568)))];
+            tensor<fp16, [1024]> var_563_to_fp16 = const()[name = tensor<string, []>("op_563_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112734784)))];
+            tensor<fp16, [1, 1500, 1024]> q_17_cast = linear(bias = var_563_to_fp16, weight = var_562_to_fp16, x = var_551_cast);
+            tensor<fp16, [1024, 1024]> var_566_to_fp16 = const()[name = tensor<string, []>("op_566_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112736896)))];
+            tensor<fp16, [1024]> k_17_bias_0_to_fp16 = const()[name = tensor<string, []>("k_17_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114834112)))];
+            tensor<fp16, [1, 1500, 1024]> k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_566_to_fp16, x = var_551_cast);
+            tensor<fp16, [1024, 1024]> var_570_to_fp16 = const()[name = tensor<string, []>("op_570_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114836224)))];
+            tensor<fp16, [1024]> var_571_to_fp16 = const()[name = tensor<string, []>("op_571_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116933440)))];
+            tensor<fp16, [1, 1500, 1024]> v_17_cast = linear(bias = var_571_to_fp16, weight = var_570_to_fp16, x = var_551_cast);
+            tensor<int32, [4]> var_579 = const()[name = tensor<string, []>("op_579"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_580_cast = reshape(shape = var_579, x = q_17_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_176_to_fp16 = const()[name = tensor<string, []>("const_176_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_19_cast = mul(x = var_580_cast, y = const_176_to_fp16);
+            tensor<int32, [4]> var_586 = const()[name = tensor<string, []>("op_586"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_587_cast = reshape(shape = var_586, x = k_17_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_177_to_fp16 = const()[name = tensor<string, []>("const_177_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_19_cast = mul(x = var_587_cast, y = const_177_to_fp16);
+            tensor<int32, [4]> var_593 = const()[name = tensor<string, []>("op_593"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_594_cast = reshape(shape = var_593, x = v_17_cast);
+            tensor<int32, [4]> var_595 = const()[name = tensor<string, []>("op_595"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_9_transpose_x_0 = const()[name = tensor<string, []>("qk_9_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_9_transpose_y_0 = const()[name = tensor<string, []>("qk_9_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_56_perm_0 = const()[name = tensor<string, []>("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_57_perm_0 = const()[name = tensor<string, []>("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_173 = transpose(perm = transpose_57_perm_0, x = k_19_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_174 = transpose(perm = transpose_56_perm_0, x = q_19_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_174, y = transpose_173);
+            tensor<fp16, [1, 16, 1500, 1500]> var_599_cast = softmax(axis = var_534, x = qk_9_cast);
+            tensor<bool, []> var_601_transpose_x_0 = const()[name = tensor<string, []>("op_601_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_601_transpose_y_0 = const()[name = tensor<string, []>("op_601_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_175 = transpose(perm = var_595, x = var_594_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_601_cast = matmul(transpose_x = var_601_transpose_x_0, transpose_y = var_601_transpose_y_0, x = var_599_cast, y = transpose_175);
+            tensor<int32, [4]> var_602 = const()[name = tensor<string, []>("op_602"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_4 = const()[name = tensor<string, []>("concat_4"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_172 = transpose(perm = var_602, x = var_601_cast);
+            tensor<fp16, [1, 1500, 1024]> x_59_cast = reshape(shape = concat_4, x = transpose_172);
+            tensor<fp16, [1024, 1024]> var_607_to_fp16 = const()[name = tensor<string, []>("op_607_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116935552)))];
+            tensor<fp16, [1024]> var_608_to_fp16 = const()[name = tensor<string, []>("op_608_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119032768)))];
+            tensor<fp16, [1, 1500, 1024]> var_609_cast = linear(bias = var_608_to_fp16, weight = var_607_to_fp16, x = x_59_cast);
+            tensor<fp16, [1, 1500, 1024]> x_61_cast = add(x = x_55_cast, y = var_609_cast);
+            tensor<int32, [1]> var_615_axes_0 = const()[name = tensor<string, []>("op_615_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119034880)))];
+            tensor<fp16, [1024]> blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119036992)))];
+            tensor<fp16, [1, 1500, 1024]> var_615_cast = layer_norm(axes = var_615_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_540_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast);
+            tensor<fp16, [4096, 1024]> var_624_to_fp16 = const()[name = tensor<string, []>("op_624_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119039104)))];
+            tensor<fp16, [4096]> var_625_to_fp16 = const()[name = tensor<string, []>("op_625_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127427776)))];
+            tensor<fp16, [1, 1500, 4096]> input_41_cast = linear(bias = var_625_to_fp16, weight = var_624_to_fp16, x = var_615_cast);
+            tensor<string, []> x_65_mode_0 = const()[name = tensor<string, []>("x_65_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast);
+            tensor<fp16, [1024, 4096]> var_630_to_fp16 = const()[name = tensor<string, []>("op_630_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127436032)))];
+            tensor<fp16, [1024]> var_631_to_fp16 = const()[name = tensor<string, []>("op_631_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135824704)))];
+            tensor<fp16, [1, 1500, 1024]> var_632_cast = linear(bias = var_631_to_fp16, weight = var_630_to_fp16, x = x_65_cast);
+            tensor<fp16, [1, 1500, 1024]> x_67_cast = add(x = x_61_cast, y = var_632_cast);
+            tensor<int32, []> var_641 = const()[name = tensor<string, []>("op_641"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_658_axes_0 = const()[name = tensor<string, []>("op_658_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135826816)))];
+            tensor<fp16, [1024]> blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135828928)))];
+            tensor<fp16, []> var_647_to_fp16 = const()[name = tensor<string, []>("op_647_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_658_cast = layer_norm(axes = var_658_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_647_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast);
+            tensor<fp16, [1024, 1024]> var_669_to_fp16 = const()[name = tensor<string, []>("op_669_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135831040)))];
+            tensor<fp16, [1024]> var_670_to_fp16 = const()[name = tensor<string, []>("op_670_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137928256)))];
+            tensor<fp16, [1, 1500, 1024]> q_21_cast = linear(bias = var_670_to_fp16, weight = var_669_to_fp16, x = var_658_cast);
+            tensor<fp16, [1024, 1024]> var_673_to_fp16 = const()[name = tensor<string, []>("op_673_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137930368)))];
+            tensor<fp16, [1024]> k_21_bias_0_to_fp16 = const()[name = tensor<string, []>("k_21_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140027584)))];
+            tensor<fp16, [1, 1500, 1024]> k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_673_to_fp16, x = var_658_cast);
+            tensor<fp16, [1024, 1024]> var_677_to_fp16 = const()[name = tensor<string, []>("op_677_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140029696)))];
+            tensor<fp16, [1024]> var_678_to_fp16 = const()[name = tensor<string, []>("op_678_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142126912)))];
+            tensor<fp16, [1, 1500, 1024]> v_21_cast = linear(bias = var_678_to_fp16, weight = var_677_to_fp16, x = var_658_cast);
+            tensor<int32, [4]> var_686 = const()[name = tensor<string, []>("op_686"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_687_cast = reshape(shape = var_686, x = q_21_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_178_to_fp16 = const()[name = tensor<string, []>("const_178_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_23_cast = mul(x = var_687_cast, y = const_178_to_fp16);
+            tensor<int32, [4]> var_693 = const()[name = tensor<string, []>("op_693"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_694_cast = reshape(shape = var_693, x = k_21_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_179_to_fp16 = const()[name = tensor<string, []>("const_179_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_23_cast = mul(x = var_694_cast, y = const_179_to_fp16);
+            tensor<int32, [4]> var_700 = const()[name = tensor<string, []>("op_700"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_701_cast = reshape(shape = var_700, x = v_21_cast);
+            tensor<int32, [4]> var_702 = const()[name = tensor<string, []>("op_702"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_11_transpose_x_0 = const()[name = tensor<string, []>("qk_11_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_11_transpose_y_0 = const()[name = tensor<string, []>("qk_11_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_58_perm_0 = const()[name = tensor<string, []>("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_59_perm_0 = const()[name = tensor<string, []>("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_169 = transpose(perm = transpose_59_perm_0, x = k_23_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_170 = transpose(perm = transpose_58_perm_0, x = q_23_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_170, y = transpose_169);
+            tensor<fp16, [1, 16, 1500, 1500]> var_706_cast = softmax(axis = var_641, x = qk_11_cast);
+            tensor<bool, []> var_708_transpose_x_0 = const()[name = tensor<string, []>("op_708_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_708_transpose_y_0 = const()[name = tensor<string, []>("op_708_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_171 = transpose(perm = var_702, x = var_701_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_708_cast = matmul(transpose_x = var_708_transpose_x_0, transpose_y = var_708_transpose_y_0, x = var_706_cast, y = transpose_171);
+            tensor<int32, [4]> var_709 = const()[name = tensor<string, []>("op_709"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_5 = const()[name = tensor<string, []>("concat_5"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_168 = transpose(perm = var_709, x = var_708_cast);
+            tensor<fp16, [1, 1500, 1024]> x_71_cast = reshape(shape = concat_5, x = transpose_168);
+            tensor<fp16, [1024, 1024]> var_714_to_fp16 = const()[name = tensor<string, []>("op_714_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142129024)))];
+            tensor<fp16, [1024]> var_715_to_fp16 = const()[name = tensor<string, []>("op_715_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144226240)))];
+            tensor<fp16, [1, 1500, 1024]> var_716_cast = linear(bias = var_715_to_fp16, weight = var_714_to_fp16, x = x_71_cast);
+            tensor<fp16, [1, 1500, 1024]> x_73_cast = add(x = x_67_cast, y = var_716_cast);
+            tensor<int32, [1]> var_722_axes_0 = const()[name = tensor<string, []>("op_722_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144228352)))];
+            tensor<fp16, [1024]> blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144230464)))];
+            tensor<fp16, [1, 1500, 1024]> var_722_cast = layer_norm(axes = var_722_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_647_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast);
+            tensor<fp16, [4096, 1024]> var_731_to_fp16 = const()[name = tensor<string, []>("op_731_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144232576)))];
+            tensor<fp16, [4096]> var_732_to_fp16 = const()[name = tensor<string, []>("op_732_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152621248)))];
+            tensor<fp16, [1, 1500, 4096]> input_49_cast = linear(bias = var_732_to_fp16, weight = var_731_to_fp16, x = var_722_cast);
+            tensor<string, []> x_77_mode_0 = const()[name = tensor<string, []>("x_77_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast);
+            tensor<fp16, [1024, 4096]> var_737_to_fp16 = const()[name = tensor<string, []>("op_737_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152629504)))];
+            tensor<fp16, [1024]> var_738_to_fp16 = const()[name = tensor<string, []>("op_738_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161018176)))];
+            tensor<fp16, [1, 1500, 1024]> var_739_cast = linear(bias = var_738_to_fp16, weight = var_737_to_fp16, x = x_77_cast);
+            tensor<fp16, [1, 1500, 1024]> x_79_cast = add(x = x_73_cast, y = var_739_cast);
+            tensor<int32, []> var_748 = const()[name = tensor<string, []>("op_748"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_765_axes_0 = const()[name = tensor<string, []>("op_765_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161020288)))];
+            tensor<fp16, [1024]> blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161022400)))];
+            tensor<fp16, []> var_754_to_fp16 = const()[name = tensor<string, []>("op_754_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_765_cast = layer_norm(axes = var_765_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_754_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast);
+            tensor<fp16, [1024, 1024]> var_776_to_fp16 = const()[name = tensor<string, []>("op_776_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161024512)))];
+            tensor<fp16, [1024]> var_777_to_fp16 = const()[name = tensor<string, []>("op_777_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163121728)))];
+            tensor<fp16, [1, 1500, 1024]> q_25_cast = linear(bias = var_777_to_fp16, weight = var_776_to_fp16, x = var_765_cast);
+            tensor<fp16, [1024, 1024]> var_780_to_fp16 = const()[name = tensor<string, []>("op_780_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163123840)))];
+            tensor<fp16, [1024]> k_25_bias_0_to_fp16 = const()[name = tensor<string, []>("k_25_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165221056)))];
+            tensor<fp16, [1, 1500, 1024]> k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_780_to_fp16, x = var_765_cast);
+            tensor<fp16, [1024, 1024]> var_784_to_fp16 = const()[name = tensor<string, []>("op_784_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165223168)))];
+            tensor<fp16, [1024]> var_785_to_fp16 = const()[name = tensor<string, []>("op_785_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167320384)))];
+            tensor<fp16, [1, 1500, 1024]> v_25_cast = linear(bias = var_785_to_fp16, weight = var_784_to_fp16, x = var_765_cast);
+            tensor<int32, [4]> var_793 = const()[name = tensor<string, []>("op_793"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_794_cast = reshape(shape = var_793, x = q_25_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_180_to_fp16 = const()[name = tensor<string, []>("const_180_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_27_cast = mul(x = var_794_cast, y = const_180_to_fp16);
+            tensor<int32, [4]> var_800 = const()[name = tensor<string, []>("op_800"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_801_cast = reshape(shape = var_800, x = k_25_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_181_to_fp16 = const()[name = tensor<string, []>("const_181_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_27_cast = mul(x = var_801_cast, y = const_181_to_fp16);
+            tensor<int32, [4]> var_807 = const()[name = tensor<string, []>("op_807"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_808_cast = reshape(shape = var_807, x = v_25_cast);
+            tensor<int32, [4]> var_809 = const()[name = tensor<string, []>("op_809"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_13_transpose_x_0 = const()[name = tensor<string, []>("qk_13_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_13_transpose_y_0 = const()[name = tensor<string, []>("qk_13_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_60_perm_0 = const()[name = tensor<string, []>("transpose_60_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_61_perm_0 = const()[name = tensor<string, []>("transpose_61_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_165 = transpose(perm = transpose_61_perm_0, x = k_27_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_166 = transpose(perm = transpose_60_perm_0, x = q_27_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_166, y = transpose_165);
+            tensor<fp16, [1, 16, 1500, 1500]> var_813_cast = softmax(axis = var_748, x = qk_13_cast);
+            tensor<bool, []> var_815_transpose_x_0 = const()[name = tensor<string, []>("op_815_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_815_transpose_y_0 = const()[name = tensor<string, []>("op_815_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_167 = transpose(perm = var_809, x = var_808_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_815_cast = matmul(transpose_x = var_815_transpose_x_0, transpose_y = var_815_transpose_y_0, x = var_813_cast, y = transpose_167);
+            tensor<int32, [4]> var_816 = const()[name = tensor<string, []>("op_816"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_6 = const()[name = tensor<string, []>("concat_6"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_164 = transpose(perm = var_816, x = var_815_cast);
+            tensor<fp16, [1, 1500, 1024]> x_83_cast = reshape(shape = concat_6, x = transpose_164);
+            tensor<fp16, [1024, 1024]> var_821_to_fp16 = const()[name = tensor<string, []>("op_821_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167322496)))];
+            tensor<fp16, [1024]> var_822_to_fp16 = const()[name = tensor<string, []>("op_822_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169419712)))];
+            tensor<fp16, [1, 1500, 1024]> var_823_cast = linear(bias = var_822_to_fp16, weight = var_821_to_fp16, x = x_83_cast);
+            tensor<fp16, [1, 1500, 1024]> x_85_cast = add(x = x_79_cast, y = var_823_cast);
+            tensor<int32, [1]> var_829_axes_0 = const()[name = tensor<string, []>("op_829_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169421824)))];
+            tensor<fp16, [1024]> blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169423936)))];
+            tensor<fp16, [1, 1500, 1024]> var_829_cast = layer_norm(axes = var_829_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_754_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast);
+            tensor<fp16, [4096, 1024]> var_838_to_fp16 = const()[name = tensor<string, []>("op_838_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169426048)))];
+            tensor<fp16, [4096]> var_839_to_fp16 = const()[name = tensor<string, []>("op_839_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177814720)))];
+            tensor<fp16, [1, 1500, 4096]> input_57_cast = linear(bias = var_839_to_fp16, weight = var_838_to_fp16, x = var_829_cast);
+            tensor<string, []> x_89_mode_0 = const()[name = tensor<string, []>("x_89_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast);
+            tensor<fp16, [1024, 4096]> var_844_to_fp16 = const()[name = tensor<string, []>("op_844_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177822976)))];
+            tensor<fp16, [1024]> var_845_to_fp16 = const()[name = tensor<string, []>("op_845_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186211648)))];
+            tensor<fp16, [1, 1500, 1024]> var_846_cast = linear(bias = var_845_to_fp16, weight = var_844_to_fp16, x = x_89_cast);
+            tensor<fp16, [1, 1500, 1024]> x_91_cast = add(x = x_85_cast, y = var_846_cast);
+            tensor<int32, []> var_855 = const()[name = tensor<string, []>("op_855"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_872_axes_0 = const()[name = tensor<string, []>("op_872_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186213760)))];
+            tensor<fp16, [1024]> blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186215872)))];
+            tensor<fp16, []> var_861_to_fp16 = const()[name = tensor<string, []>("op_861_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_872_cast = layer_norm(axes = var_872_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_861_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast);
+            tensor<fp16, [1024, 1024]> var_883_to_fp16 = const()[name = tensor<string, []>("op_883_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186217984)))];
+            tensor<fp16, [1024]> var_884_to_fp16 = const()[name = tensor<string, []>("op_884_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188315200)))];
+            tensor<fp16, [1, 1500, 1024]> q_29_cast = linear(bias = var_884_to_fp16, weight = var_883_to_fp16, x = var_872_cast);
+            tensor<fp16, [1024, 1024]> var_887_to_fp16 = const()[name = tensor<string, []>("op_887_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188317312)))];
+            tensor<fp16, [1024]> k_29_bias_0_to_fp16 = const()[name = tensor<string, []>("k_29_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190414528)))];
+            tensor<fp16, [1, 1500, 1024]> k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_887_to_fp16, x = var_872_cast);
+            tensor<fp16, [1024, 1024]> var_891_to_fp16 = const()[name = tensor<string, []>("op_891_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190416640)))];
+            tensor<fp16, [1024]> var_892_to_fp16 = const()[name = tensor<string, []>("op_892_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192513856)))];
+            tensor<fp16, [1, 1500, 1024]> v_29_cast = linear(bias = var_892_to_fp16, weight = var_891_to_fp16, x = var_872_cast);
+            tensor<int32, [4]> var_900 = const()[name = tensor<string, []>("op_900"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_901_cast = reshape(shape = var_900, x = q_29_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_182_to_fp16 = const()[name = tensor<string, []>("const_182_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_31_cast = mul(x = var_901_cast, y = const_182_to_fp16);
+            tensor<int32, [4]> var_907 = const()[name = tensor<string, []>("op_907"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_908_cast = reshape(shape = var_907, x = k_29_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_183_to_fp16 = const()[name = tensor<string, []>("const_183_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_31_cast = mul(x = var_908_cast, y = const_183_to_fp16);
+            tensor<int32, [4]> var_914 = const()[name = tensor<string, []>("op_914"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_915_cast = reshape(shape = var_914, x = v_29_cast);
+            tensor<int32, [4]> var_916 = const()[name = tensor<string, []>("op_916"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_15_transpose_x_0 = const()[name = tensor<string, []>("qk_15_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_15_transpose_y_0 = const()[name = tensor<string, []>("qk_15_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_62_perm_0 = const()[name = tensor<string, []>("transpose_62_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_63_perm_0 = const()[name = tensor<string, []>("transpose_63_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_161 = transpose(perm = transpose_63_perm_0, x = k_31_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_162 = transpose(perm = transpose_62_perm_0, x = q_31_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_162, y = transpose_161);
+            tensor<fp16, [1, 16, 1500, 1500]> var_920_cast = softmax(axis = var_855, x = qk_15_cast);
+            tensor<bool, []> var_922_transpose_x_0 = const()[name = tensor<string, []>("op_922_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_922_transpose_y_0 = const()[name = tensor<string, []>("op_922_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_163 = transpose(perm = var_916, x = var_915_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_922_cast = matmul(transpose_x = var_922_transpose_x_0, transpose_y = var_922_transpose_y_0, x = var_920_cast, y = transpose_163);
+            tensor<int32, [4]> var_923 = const()[name = tensor<string, []>("op_923"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_7 = const()[name = tensor<string, []>("concat_7"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_160 = transpose(perm = var_923, x = var_922_cast);
+            tensor<fp16, [1, 1500, 1024]> x_95_cast = reshape(shape = concat_7, x = transpose_160);
+            tensor<fp16, [1024, 1024]> var_928_to_fp16 = const()[name = tensor<string, []>("op_928_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192515968)))];
+            tensor<fp16, [1024]> var_929_to_fp16 = const()[name = tensor<string, []>("op_929_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194613184)))];
+            tensor<fp16, [1, 1500, 1024]> var_930_cast = linear(bias = var_929_to_fp16, weight = var_928_to_fp16, x = x_95_cast);
+            tensor<fp16, [1, 1500, 1024]> x_97_cast = add(x = x_91_cast, y = var_930_cast);
+            tensor<int32, [1]> var_936_axes_0 = const()[name = tensor<string, []>("op_936_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194615296)))];
+            tensor<fp16, [1024]> blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194617408)))];
+            tensor<fp16, [1, 1500, 1024]> var_936_cast = layer_norm(axes = var_936_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_861_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast);
+            tensor<fp16, [4096, 1024]> var_945_to_fp16 = const()[name = tensor<string, []>("op_945_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194619520)))];
+            tensor<fp16, [4096]> var_946_to_fp16 = const()[name = tensor<string, []>("op_946_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203008192)))];
+            tensor<fp16, [1, 1500, 4096]> input_65_cast = linear(bias = var_946_to_fp16, weight = var_945_to_fp16, x = var_936_cast);
+            tensor<string, []> x_101_mode_0 = const()[name = tensor<string, []>("x_101_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast);
+            tensor<fp16, [1024, 4096]> var_951_to_fp16 = const()[name = tensor<string, []>("op_951_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203016448)))];
+            tensor<fp16, [1024]> var_952_to_fp16 = const()[name = tensor<string, []>("op_952_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211405120)))];
+            tensor<fp16, [1, 1500, 1024]> var_953_cast = linear(bias = var_952_to_fp16, weight = var_951_to_fp16, x = x_101_cast);
+            tensor<fp16, [1, 1500, 1024]> x_103_cast = add(x = x_97_cast, y = var_953_cast);
+            tensor<int32, []> var_962 = const()[name = tensor<string, []>("op_962"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_979_axes_0 = const()[name = tensor<string, []>("op_979_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211407232)))];
+            tensor<fp16, [1024]> blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211409344)))];
+            tensor<fp16, []> var_968_to_fp16 = const()[name = tensor<string, []>("op_968_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_979_cast = layer_norm(axes = var_979_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_968_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast);
+            tensor<fp16, [1024, 1024]> var_990_to_fp16 = const()[name = tensor<string, []>("op_990_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211411456)))];
+            tensor<fp16, [1024]> var_991_to_fp16 = const()[name = tensor<string, []>("op_991_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213508672)))];
+            tensor<fp16, [1, 1500, 1024]> q_33_cast = linear(bias = var_991_to_fp16, weight = var_990_to_fp16, x = var_979_cast);
+            tensor<fp16, [1024, 1024]> var_994_to_fp16 = const()[name = tensor<string, []>("op_994_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213510784)))];
+            tensor<fp16, [1024]> k_33_bias_0_to_fp16 = const()[name = tensor<string, []>("k_33_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215608000)))];
+            tensor<fp16, [1, 1500, 1024]> k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_994_to_fp16, x = var_979_cast);
+            tensor<fp16, [1024, 1024]> var_998_to_fp16 = const()[name = tensor<string, []>("op_998_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215610112)))];
+            tensor<fp16, [1024]> var_999_to_fp16 = const()[name = tensor<string, []>("op_999_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217707328)))];
+            tensor<fp16, [1, 1500, 1024]> v_33_cast = linear(bias = var_999_to_fp16, weight = var_998_to_fp16, x = var_979_cast);
+            tensor<int32, [4]> var_1007 = const()[name = tensor<string, []>("op_1007"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1008_cast = reshape(shape = var_1007, x = q_33_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_184_to_fp16 = const()[name = tensor<string, []>("const_184_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_35_cast = mul(x = var_1008_cast, y = const_184_to_fp16);
+            tensor<int32, [4]> var_1014 = const()[name = tensor<string, []>("op_1014"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1015_cast = reshape(shape = var_1014, x = k_33_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_185_to_fp16 = const()[name = tensor<string, []>("const_185_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_35_cast = mul(x = var_1015_cast, y = const_185_to_fp16);
+            tensor<int32, [4]> var_1021 = const()[name = tensor<string, []>("op_1021"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1022_cast = reshape(shape = var_1021, x = v_33_cast);
+            tensor<int32, [4]> var_1023 = const()[name = tensor<string, []>("op_1023"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_17_transpose_x_0 = const()[name = tensor<string, []>("qk_17_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_17_transpose_y_0 = const()[name = tensor<string, []>("qk_17_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_64_perm_0 = const()[name = tensor<string, []>("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_65_perm_0 = const()[name = tensor<string, []>("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_157 = transpose(perm = transpose_65_perm_0, x = k_35_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_158 = transpose(perm = transpose_64_perm_0, x = q_35_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_158, y = transpose_157);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1027_cast = softmax(axis = var_962, x = qk_17_cast);
+            tensor<bool, []> var_1029_transpose_x_0 = const()[name = tensor<string, []>("op_1029_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1029_transpose_y_0 = const()[name = tensor<string, []>("op_1029_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_159 = transpose(perm = var_1023, x = var_1022_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1029_cast = matmul(transpose_x = var_1029_transpose_x_0, transpose_y = var_1029_transpose_y_0, x = var_1027_cast, y = transpose_159);
+            tensor<int32, [4]> var_1030 = const()[name = tensor<string, []>("op_1030"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_8 = const()[name = tensor<string, []>("concat_8"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_156 = transpose(perm = var_1030, x = var_1029_cast);
+            tensor<fp16, [1, 1500, 1024]> x_107_cast = reshape(shape = concat_8, x = transpose_156);
+            tensor<fp16, [1024, 1024]> var_1035_to_fp16 = const()[name = tensor<string, []>("op_1035_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217709440)))];
+            tensor<fp16, [1024]> var_1036_to_fp16 = const()[name = tensor<string, []>("op_1036_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219806656)))];
+            tensor<fp16, [1, 1500, 1024]> var_1037_cast = linear(bias = var_1036_to_fp16, weight = var_1035_to_fp16, x = x_107_cast);
+            tensor<fp16, [1, 1500, 1024]> x_109_cast = add(x = x_103_cast, y = var_1037_cast);
+            tensor<int32, [1]> var_1043_axes_0 = const()[name = tensor<string, []>("op_1043_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219808768)))];
+            tensor<fp16, [1024]> blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219810880)))];
+            tensor<fp16, [1, 1500, 1024]> var_1043_cast = layer_norm(axes = var_1043_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_968_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast);
+            tensor<fp16, [4096, 1024]> var_1052_to_fp16 = const()[name = tensor<string, []>("op_1052_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219812992)))];
+            tensor<fp16, [4096]> var_1053_to_fp16 = const()[name = tensor<string, []>("op_1053_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228201664)))];
+            tensor<fp16, [1, 1500, 4096]> input_73_cast = linear(bias = var_1053_to_fp16, weight = var_1052_to_fp16, x = var_1043_cast);
+            tensor<string, []> x_113_mode_0 = const()[name = tensor<string, []>("x_113_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast);
+            tensor<fp16, [1024, 4096]> var_1058_to_fp16 = const()[name = tensor<string, []>("op_1058_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228209920)))];
+            tensor<fp16, [1024]> var_1059_to_fp16 = const()[name = tensor<string, []>("op_1059_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236598592)))];
+            tensor<fp16, [1, 1500, 1024]> var_1060_cast = linear(bias = var_1059_to_fp16, weight = var_1058_to_fp16, x = x_113_cast);
+            tensor<fp16, [1, 1500, 1024]> x_115_cast = add(x = x_109_cast, y = var_1060_cast);
+            tensor<int32, []> var_1069 = const()[name = tensor<string, []>("op_1069"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1086_axes_0 = const()[name = tensor<string, []>("op_1086_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236600704)))];
+            tensor<fp16, [1024]> blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236602816)))];
+            tensor<fp16, []> var_1075_to_fp16 = const()[name = tensor<string, []>("op_1075_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1086_cast = layer_norm(axes = var_1086_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1075_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast);
+            tensor<fp16, [1024, 1024]> var_1097_to_fp16 = const()[name = tensor<string, []>("op_1097_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236604928)))];
+            tensor<fp16, [1024]> var_1098_to_fp16 = const()[name = tensor<string, []>("op_1098_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238702144)))];
+            tensor<fp16, [1, 1500, 1024]> q_37_cast = linear(bias = var_1098_to_fp16, weight = var_1097_to_fp16, x = var_1086_cast);
+            tensor<fp16, [1024, 1024]> var_1101_to_fp16 = const()[name = tensor<string, []>("op_1101_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238704256)))];
+            tensor<fp16, [1024]> k_37_bias_0_to_fp16 = const()[name = tensor<string, []>("k_37_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(240801472)))];
+            tensor<fp16, [1, 1500, 1024]> k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1101_to_fp16, x = var_1086_cast);
+            tensor<fp16, [1024, 1024]> var_1105_to_fp16 = const()[name = tensor<string, []>("op_1105_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(240803584)))];
+            tensor<fp16, [1024]> var_1106_to_fp16 = const()[name = tensor<string, []>("op_1106_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(242900800)))];
+            tensor<fp16, [1, 1500, 1024]> v_37_cast = linear(bias = var_1106_to_fp16, weight = var_1105_to_fp16, x = var_1086_cast);
+            tensor<int32, [4]> var_1114 = const()[name = tensor<string, []>("op_1114"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1115_cast = reshape(shape = var_1114, x = q_37_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_186_to_fp16 = const()[name = tensor<string, []>("const_186_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_39_cast = mul(x = var_1115_cast, y = const_186_to_fp16);
+            tensor<int32, [4]> var_1121 = const()[name = tensor<string, []>("op_1121"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1122_cast = reshape(shape = var_1121, x = k_37_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_187_to_fp16 = const()[name = tensor<string, []>("const_187_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_39_cast = mul(x = var_1122_cast, y = const_187_to_fp16);
+            tensor<int32, [4]> var_1128 = const()[name = tensor<string, []>("op_1128"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1129_cast = reshape(shape = var_1128, x = v_37_cast);
+            tensor<int32, [4]> var_1130 = const()[name = tensor<string, []>("op_1130"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_19_transpose_x_0 = const()[name = tensor<string, []>("qk_19_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_19_transpose_y_0 = const()[name = tensor<string, []>("qk_19_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_66_perm_0 = const()[name = tensor<string, []>("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_67_perm_0 = const()[name = tensor<string, []>("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_153 = transpose(perm = transpose_67_perm_0, x = k_39_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_154 = transpose(perm = transpose_66_perm_0, x = q_39_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_154, y = transpose_153);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1134_cast = softmax(axis = var_1069, x = qk_19_cast);
+            tensor<bool, []> var_1136_transpose_x_0 = const()[name = tensor<string, []>("op_1136_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1136_transpose_y_0 = const()[name = tensor<string, []>("op_1136_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_155 = transpose(perm = var_1130, x = var_1129_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1136_cast = matmul(transpose_x = var_1136_transpose_x_0, transpose_y = var_1136_transpose_y_0, x = var_1134_cast, y = transpose_155);
+            tensor<int32, [4]> var_1137 = const()[name = tensor<string, []>("op_1137"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_9 = const()[name = tensor<string, []>("concat_9"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_152 = transpose(perm = var_1137, x = var_1136_cast);
+            tensor<fp16, [1, 1500, 1024]> x_119_cast = reshape(shape = concat_9, x = transpose_152);
+            tensor<fp16, [1024, 1024]> var_1142_to_fp16 = const()[name = tensor<string, []>("op_1142_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(242902912)))];
+            tensor<fp16, [1024]> var_1143_to_fp16 = const()[name = tensor<string, []>("op_1143_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245000128)))];
+            tensor<fp16, [1, 1500, 1024]> var_1144_cast = linear(bias = var_1143_to_fp16, weight = var_1142_to_fp16, x = x_119_cast);
+            tensor<fp16, [1, 1500, 1024]> x_121_cast = add(x = x_115_cast, y = var_1144_cast);
+            tensor<int32, [1]> var_1150_axes_0 = const()[name = tensor<string, []>("op_1150_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245002240)))];
+            tensor<fp16, [1024]> blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245004352)))];
+            tensor<fp16, [1, 1500, 1024]> var_1150_cast = layer_norm(axes = var_1150_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1075_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast);
+            tensor<fp16, [4096, 1024]> var_1159_to_fp16 = const()[name = tensor<string, []>("op_1159_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245006464)))];
+            tensor<fp16, [4096]> var_1160_to_fp16 = const()[name = tensor<string, []>("op_1160_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253395136)))];
+            tensor<fp16, [1, 1500, 4096]> input_81_cast = linear(bias = var_1160_to_fp16, weight = var_1159_to_fp16, x = var_1150_cast);
+            tensor<string, []> x_125_mode_0 = const()[name = tensor<string, []>("x_125_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast);
+            tensor<fp16, [1024, 4096]> var_1165_to_fp16 = const()[name = tensor<string, []>("op_1165_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253403392)))];
+            tensor<fp16, [1024]> var_1166_to_fp16 = const()[name = tensor<string, []>("op_1166_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261792064)))];
+            tensor<fp16, [1, 1500, 1024]> var_1167_cast = linear(bias = var_1166_to_fp16, weight = var_1165_to_fp16, x = x_125_cast);
+            tensor<fp16, [1, 1500, 1024]> x_127_cast = add(x = x_121_cast, y = var_1167_cast);
+            tensor<int32, []> var_1176 = const()[name = tensor<string, []>("op_1176"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1193_axes_0 = const()[name = tensor<string, []>("op_1193_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261794176)))];
+            tensor<fp16, [1024]> blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261796288)))];
+            tensor<fp16, []> var_1182_to_fp16 = const()[name = tensor<string, []>("op_1182_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1193_cast = layer_norm(axes = var_1193_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1182_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast);
+            tensor<fp16, [1024, 1024]> var_1204_to_fp16 = const()[name = tensor<string, []>("op_1204_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261798400)))];
+            tensor<fp16, [1024]> var_1205_to_fp16 = const()[name = tensor<string, []>("op_1205_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263895616)))];
+            tensor<fp16, [1, 1500, 1024]> q_41_cast = linear(bias = var_1205_to_fp16, weight = var_1204_to_fp16, x = var_1193_cast);
+            tensor<fp16, [1024, 1024]> var_1208_to_fp16 = const()[name = tensor<string, []>("op_1208_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263897728)))];
+            tensor<fp16, [1024]> k_41_bias_0_to_fp16 = const()[name = tensor<string, []>("k_41_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265994944)))];
+            tensor<fp16, [1, 1500, 1024]> k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1208_to_fp16, x = var_1193_cast);
+            tensor<fp16, [1024, 1024]> var_1212_to_fp16 = const()[name = tensor<string, []>("op_1212_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265997056)))];
+            tensor<fp16, [1024]> var_1213_to_fp16 = const()[name = tensor<string, []>("op_1213_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(268094272)))];
+            tensor<fp16, [1, 1500, 1024]> v_41_cast = linear(bias = var_1213_to_fp16, weight = var_1212_to_fp16, x = var_1193_cast);
+            tensor<int32, [4]> var_1221 = const()[name = tensor<string, []>("op_1221"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1222_cast = reshape(shape = var_1221, x = q_41_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_188_to_fp16 = const()[name = tensor<string, []>("const_188_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_43_cast = mul(x = var_1222_cast, y = const_188_to_fp16);
+            tensor<int32, [4]> var_1228 = const()[name = tensor<string, []>("op_1228"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1229_cast = reshape(shape = var_1228, x = k_41_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_189_to_fp16 = const()[name = tensor<string, []>("const_189_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_43_cast = mul(x = var_1229_cast, y = const_189_to_fp16);
+            tensor<int32, [4]> var_1235 = const()[name = tensor<string, []>("op_1235"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1236_cast = reshape(shape = var_1235, x = v_41_cast);
+            tensor<int32, [4]> var_1237 = const()[name = tensor<string, []>("op_1237"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_21_transpose_x_0 = const()[name = tensor<string, []>("qk_21_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_21_transpose_y_0 = const()[name = tensor<string, []>("qk_21_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_68_perm_0 = const()[name = tensor<string, []>("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_69_perm_0 = const()[name = tensor<string, []>("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_149 = transpose(perm = transpose_69_perm_0, x = k_43_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_150 = transpose(perm = transpose_68_perm_0, x = q_43_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_150, y = transpose_149);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1241_cast = softmax(axis = var_1176, x = qk_21_cast);
+            tensor<bool, []> var_1243_transpose_x_0 = const()[name = tensor<string, []>("op_1243_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1243_transpose_y_0 = const()[name = tensor<string, []>("op_1243_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_151 = transpose(perm = var_1237, x = var_1236_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1243_cast = matmul(transpose_x = var_1243_transpose_x_0, transpose_y = var_1243_transpose_y_0, x = var_1241_cast, y = transpose_151);
+            tensor<int32, [4]> var_1244 = const()[name = tensor<string, []>("op_1244"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_10 = const()[name = tensor<string, []>("concat_10"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_148 = transpose(perm = var_1244, x = var_1243_cast);
+            tensor<fp16, [1, 1500, 1024]> x_131_cast = reshape(shape = concat_10, x = transpose_148);
+            tensor<fp16, [1024, 1024]> var_1249_to_fp16 = const()[name = tensor<string, []>("op_1249_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(268096384)))];
+            tensor<fp16, [1024]> var_1250_to_fp16 = const()[name = tensor<string, []>("op_1250_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270193600)))];
+            tensor<fp16, [1, 1500, 1024]> var_1251_cast = linear(bias = var_1250_to_fp16, weight = var_1249_to_fp16, x = x_131_cast);
+            tensor<fp16, [1, 1500, 1024]> x_133_cast = add(x = x_127_cast, y = var_1251_cast);
+            tensor<int32, [1]> var_1257_axes_0 = const()[name = tensor<string, []>("op_1257_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270195712)))];
+            tensor<fp16, [1024]> blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270197824)))];
+            tensor<fp16, [1, 1500, 1024]> var_1257_cast = layer_norm(axes = var_1257_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1182_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast);
+            tensor<fp16, [4096, 1024]> var_1266_to_fp16 = const()[name = tensor<string, []>("op_1266_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270199936)))];
+            tensor<fp16, [4096]> var_1267_to_fp16 = const()[name = tensor<string, []>("op_1267_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278588608)))];
+            tensor<fp16, [1, 1500, 4096]> input_89_cast = linear(bias = var_1267_to_fp16, weight = var_1266_to_fp16, x = var_1257_cast);
+            tensor<string, []> x_137_mode_0 = const()[name = tensor<string, []>("x_137_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast);
+            tensor<fp16, [1024, 4096]> var_1272_to_fp16 = const()[name = tensor<string, []>("op_1272_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278596864)))];
+            tensor<fp16, [1024]> var_1273_to_fp16 = const()[name = tensor<string, []>("op_1273_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286985536)))];
+            tensor<fp16, [1, 1500, 1024]> var_1274_cast = linear(bias = var_1273_to_fp16, weight = var_1272_to_fp16, x = x_137_cast);
+            tensor<fp16, [1, 1500, 1024]> x_139_cast = add(x = x_133_cast, y = var_1274_cast);
+            tensor<int32, []> var_1283 = const()[name = tensor<string, []>("op_1283"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1300_axes_0 = const()[name = tensor<string, []>("op_1300_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286987648)))];
+            tensor<fp16, [1024]> blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286989760)))];
+            tensor<fp16, []> var_1289_to_fp16 = const()[name = tensor<string, []>("op_1289_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1300_cast = layer_norm(axes = var_1300_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1289_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast);
+            tensor<fp16, [1024, 1024]> var_1311_to_fp16 = const()[name = tensor<string, []>("op_1311_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(286991872)))];
+            tensor<fp16, [1024]> var_1312_to_fp16 = const()[name = tensor<string, []>("op_1312_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289089088)))];
+            tensor<fp16, [1, 1500, 1024]> q_45_cast = linear(bias = var_1312_to_fp16, weight = var_1311_to_fp16, x = var_1300_cast);
+            tensor<fp16, [1024, 1024]> var_1315_to_fp16 = const()[name = tensor<string, []>("op_1315_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289091200)))];
+            tensor<fp16, [1024]> k_45_bias_0_to_fp16 = const()[name = tensor<string, []>("k_45_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291188416)))];
+            tensor<fp16, [1, 1500, 1024]> k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1315_to_fp16, x = var_1300_cast);
+            tensor<fp16, [1024, 1024]> var_1319_to_fp16 = const()[name = tensor<string, []>("op_1319_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291190528)))];
+            tensor<fp16, [1024]> var_1320_to_fp16 = const()[name = tensor<string, []>("op_1320_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293287744)))];
+            tensor<fp16, [1, 1500, 1024]> v_45_cast = linear(bias = var_1320_to_fp16, weight = var_1319_to_fp16, x = var_1300_cast);
+            tensor<int32, [4]> var_1328 = const()[name = tensor<string, []>("op_1328"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1329_cast = reshape(shape = var_1328, x = q_45_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_190_to_fp16 = const()[name = tensor<string, []>("const_190_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_47_cast = mul(x = var_1329_cast, y = const_190_to_fp16);
+            tensor<int32, [4]> var_1335 = const()[name = tensor<string, []>("op_1335"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1336_cast = reshape(shape = var_1335, x = k_45_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_191_to_fp16 = const()[name = tensor<string, []>("const_191_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_47_cast = mul(x = var_1336_cast, y = const_191_to_fp16);
+            tensor<int32, [4]> var_1342 = const()[name = tensor<string, []>("op_1342"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1343_cast = reshape(shape = var_1342, x = v_45_cast);
+            tensor<int32, [4]> var_1344 = const()[name = tensor<string, []>("op_1344"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_23_transpose_x_0 = const()[name = tensor<string, []>("qk_23_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_23_transpose_y_0 = const()[name = tensor<string, []>("qk_23_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_70_perm_0 = const()[name = tensor<string, []>("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_71_perm_0 = const()[name = tensor<string, []>("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_145 = transpose(perm = transpose_71_perm_0, x = k_47_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_146 = transpose(perm = transpose_70_perm_0, x = q_47_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_23_cast = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_146, y = transpose_145);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1348_cast = softmax(axis = var_1283, x = qk_23_cast);
+            tensor<bool, []> var_1350_transpose_x_0 = const()[name = tensor<string, []>("op_1350_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1350_transpose_y_0 = const()[name = tensor<string, []>("op_1350_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_147 = transpose(perm = var_1344, x = var_1343_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1350_cast = matmul(transpose_x = var_1350_transpose_x_0, transpose_y = var_1350_transpose_y_0, x = var_1348_cast, y = transpose_147);
+            tensor<int32, [4]> var_1351 = const()[name = tensor<string, []>("op_1351"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_11 = const()[name = tensor<string, []>("concat_11"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_144 = transpose(perm = var_1351, x = var_1350_cast);
+            tensor<fp16, [1, 1500, 1024]> x_143_cast = reshape(shape = concat_11, x = transpose_144);
+            tensor<fp16, [1024, 1024]> var_1356_to_fp16 = const()[name = tensor<string, []>("op_1356_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293289856)))];
+            tensor<fp16, [1024]> var_1357_to_fp16 = const()[name = tensor<string, []>("op_1357_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295387072)))];
+            tensor<fp16, [1, 1500, 1024]> var_1358_cast = linear(bias = var_1357_to_fp16, weight = var_1356_to_fp16, x = x_143_cast);
+            tensor<fp16, [1, 1500, 1024]> x_145_cast = add(x = x_139_cast, y = var_1358_cast);
+            tensor<int32, [1]> var_1364_axes_0 = const()[name = tensor<string, []>("op_1364_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295389184)))];
+            tensor<fp16, [1024]> blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295391296)))];
+            tensor<fp16, [1, 1500, 1024]> var_1364_cast = layer_norm(axes = var_1364_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1289_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast);
+            tensor<fp16, [4096, 1024]> var_1373_to_fp16 = const()[name = tensor<string, []>("op_1373_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295393408)))];
+            tensor<fp16, [4096]> var_1374_to_fp16 = const()[name = tensor<string, []>("op_1374_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303782080)))];
+            tensor<fp16, [1, 1500, 4096]> input_97_cast = linear(bias = var_1374_to_fp16, weight = var_1373_to_fp16, x = var_1364_cast);
+            tensor<string, []> x_149_mode_0 = const()[name = tensor<string, []>("x_149_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast);
+            tensor<fp16, [1024, 4096]> var_1379_to_fp16 = const()[name = tensor<string, []>("op_1379_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303790336)))];
+            tensor<fp16, [1024]> var_1380_to_fp16 = const()[name = tensor<string, []>("op_1380_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312179008)))];
+            tensor<fp16, [1, 1500, 1024]> var_1381_cast = linear(bias = var_1380_to_fp16, weight = var_1379_to_fp16, x = x_149_cast);
+            tensor<fp16, [1, 1500, 1024]> x_151_cast = add(x = x_145_cast, y = var_1381_cast);
+            tensor<int32, []> var_1390 = const()[name = tensor<string, []>("op_1390"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1407_axes_0 = const()[name = tensor<string, []>("op_1407_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_12_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312181120)))];
+            tensor<fp16, [1024]> blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312183232)))];
+            tensor<fp16, []> var_1396_to_fp16 = const()[name = tensor<string, []>("op_1396_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1407_cast = layer_norm(axes = var_1407_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1396_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast);
+            tensor<fp16, [1024, 1024]> var_1418_to_fp16 = const()[name = tensor<string, []>("op_1418_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312185344)))];
+            tensor<fp16, [1024]> var_1419_to_fp16 = const()[name = tensor<string, []>("op_1419_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(314282560)))];
+            tensor<fp16, [1, 1500, 1024]> q_49_cast = linear(bias = var_1419_to_fp16, weight = var_1418_to_fp16, x = var_1407_cast);
+            tensor<fp16, [1024, 1024]> var_1422_to_fp16 = const()[name = tensor<string, []>("op_1422_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(314284672)))];
+            tensor<fp16, [1024]> k_49_bias_0_to_fp16 = const()[name = tensor<string, []>("k_49_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316381888)))];
+            tensor<fp16, [1, 1500, 1024]> k_49_cast = linear(bias = k_49_bias_0_to_fp16, weight = var_1422_to_fp16, x = var_1407_cast);
+            tensor<fp16, [1024, 1024]> var_1426_to_fp16 = const()[name = tensor<string, []>("op_1426_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316384000)))];
+            tensor<fp16, [1024]> var_1427_to_fp16 = const()[name = tensor<string, []>("op_1427_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318481216)))];
+            tensor<fp16, [1, 1500, 1024]> v_49_cast = linear(bias = var_1427_to_fp16, weight = var_1426_to_fp16, x = var_1407_cast);
+            tensor<int32, [4]> var_1435 = const()[name = tensor<string, []>("op_1435"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1436_cast = reshape(shape = var_1435, x = q_49_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_192_to_fp16 = const()[name = tensor<string, []>("const_192_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_51_cast = mul(x = var_1436_cast, y = const_192_to_fp16);
+            tensor<int32, [4]> var_1442 = const()[name = tensor<string, []>("op_1442"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1443_cast = reshape(shape = var_1442, x = k_49_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_193_to_fp16 = const()[name = tensor<string, []>("const_193_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_51_cast = mul(x = var_1443_cast, y = const_193_to_fp16);
+            tensor<int32, [4]> var_1449 = const()[name = tensor<string, []>("op_1449"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1450_cast = reshape(shape = var_1449, x = v_49_cast);
+            tensor<int32, [4]> var_1451 = const()[name = tensor<string, []>("op_1451"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_25_transpose_x_0 = const()[name = tensor<string, []>("qk_25_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_25_transpose_y_0 = const()[name = tensor<string, []>("qk_25_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_72_perm_0 = const()[name = tensor<string, []>("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_73_perm_0 = const()[name = tensor<string, []>("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_141 = transpose(perm = transpose_73_perm_0, x = k_51_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_142 = transpose(perm = transpose_72_perm_0, x = q_51_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_25_cast = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_142, y = transpose_141);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1455_cast = softmax(axis = var_1390, x = qk_25_cast);
+            tensor<bool, []> var_1457_transpose_x_0 = const()[name = tensor<string, []>("op_1457_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1457_transpose_y_0 = const()[name = tensor<string, []>("op_1457_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_143 = transpose(perm = var_1451, x = var_1450_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1457_cast = matmul(transpose_x = var_1457_transpose_x_0, transpose_y = var_1457_transpose_y_0, x = var_1455_cast, y = transpose_143);
+            tensor<int32, [4]> var_1458 = const()[name = tensor<string, []>("op_1458"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_12 = const()[name = tensor<string, []>("concat_12"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_140 = transpose(perm = var_1458, x = var_1457_cast);
+            tensor<fp16, [1, 1500, 1024]> x_155_cast = reshape(shape = concat_12, x = transpose_140);
+            tensor<fp16, [1024, 1024]> var_1463_to_fp16 = const()[name = tensor<string, []>("op_1463_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318483328)))];
+            tensor<fp16, [1024]> var_1464_to_fp16 = const()[name = tensor<string, []>("op_1464_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320580544)))];
+            tensor<fp16, [1, 1500, 1024]> var_1465_cast = linear(bias = var_1464_to_fp16, weight = var_1463_to_fp16, x = x_155_cast);
+            tensor<fp16, [1, 1500, 1024]> x_157_cast = add(x = x_151_cast, y = var_1465_cast);
+            tensor<int32, [1]> var_1471_axes_0 = const()[name = tensor<string, []>("op_1471_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_12_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320582656)))];
+            tensor<fp16, [1024]> blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320584768)))];
+            tensor<fp16, [1, 1500, 1024]> var_1471_cast = layer_norm(axes = var_1471_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1396_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast);
+            tensor<fp16, [4096, 1024]> var_1480_to_fp16 = const()[name = tensor<string, []>("op_1480_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320586880)))];
+            tensor<fp16, [4096]> var_1481_to_fp16 = const()[name = tensor<string, []>("op_1481_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328975552)))];
+            tensor<fp16, [1, 1500, 4096]> input_105_cast = linear(bias = var_1481_to_fp16, weight = var_1480_to_fp16, x = var_1471_cast);
+            tensor<string, []> x_161_mode_0 = const()[name = tensor<string, []>("x_161_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_161_cast = gelu(mode = x_161_mode_0, x = input_105_cast);
+            tensor<fp16, [1024, 4096]> var_1486_to_fp16 = const()[name = tensor<string, []>("op_1486_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328983808)))];
+            tensor<fp16, [1024]> var_1487_to_fp16 = const()[name = tensor<string, []>("op_1487_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337372480)))];
+            tensor<fp16, [1, 1500, 1024]> var_1488_cast = linear(bias = var_1487_to_fp16, weight = var_1486_to_fp16, x = x_161_cast);
+            tensor<fp16, [1, 1500, 1024]> x_163_cast = add(x = x_157_cast, y = var_1488_cast);
+            tensor<int32, []> var_1497 = const()[name = tensor<string, []>("op_1497"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1514_axes_0 = const()[name = tensor<string, []>("op_1514_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_13_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337374592)))];
+            tensor<fp16, [1024]> blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337376704)))];
+            tensor<fp16, []> var_1503_to_fp16 = const()[name = tensor<string, []>("op_1503_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1514_cast = layer_norm(axes = var_1514_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1503_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast);
+            tensor<fp16, [1024, 1024]> var_1525_to_fp16 = const()[name = tensor<string, []>("op_1525_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337378816)))];
+            tensor<fp16, [1024]> var_1526_to_fp16 = const()[name = tensor<string, []>("op_1526_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339476032)))];
+            tensor<fp16, [1, 1500, 1024]> q_53_cast = linear(bias = var_1526_to_fp16, weight = var_1525_to_fp16, x = var_1514_cast);
+            tensor<fp16, [1024, 1024]> var_1529_to_fp16 = const()[name = tensor<string, []>("op_1529_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339478144)))];
+            tensor<fp16, [1024]> k_53_bias_0_to_fp16 = const()[name = tensor<string, []>("k_53_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341575360)))];
+            tensor<fp16, [1, 1500, 1024]> k_53_cast = linear(bias = k_53_bias_0_to_fp16, weight = var_1529_to_fp16, x = var_1514_cast);
+            tensor<fp16, [1024, 1024]> var_1533_to_fp16 = const()[name = tensor<string, []>("op_1533_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341577472)))];
+            tensor<fp16, [1024]> var_1534_to_fp16 = const()[name = tensor<string, []>("op_1534_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343674688)))];
+            tensor<fp16, [1, 1500, 1024]> v_53_cast = linear(bias = var_1534_to_fp16, weight = var_1533_to_fp16, x = var_1514_cast);
+            tensor<int32, [4]> var_1542 = const()[name = tensor<string, []>("op_1542"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1543_cast = reshape(shape = var_1542, x = q_53_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_194_to_fp16 = const()[name = tensor<string, []>("const_194_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_55_cast = mul(x = var_1543_cast, y = const_194_to_fp16);
+            tensor<int32, [4]> var_1549 = const()[name = tensor<string, []>("op_1549"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1550_cast = reshape(shape = var_1549, x = k_53_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_195_to_fp16 = const()[name = tensor<string, []>("const_195_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_55_cast = mul(x = var_1550_cast, y = const_195_to_fp16);
+            tensor<int32, [4]> var_1556 = const()[name = tensor<string, []>("op_1556"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1557_cast = reshape(shape = var_1556, x = v_53_cast);
+            tensor<int32, [4]> var_1558 = const()[name = tensor<string, []>("op_1558"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_27_transpose_x_0 = const()[name = tensor<string, []>("qk_27_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_27_transpose_y_0 = const()[name = tensor<string, []>("qk_27_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_74_perm_0 = const()[name = tensor<string, []>("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_75_perm_0 = const()[name = tensor<string, []>("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_137 = transpose(perm = transpose_75_perm_0, x = k_55_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_138 = transpose(perm = transpose_74_perm_0, x = q_55_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_27_cast = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_138, y = transpose_137);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1562_cast = softmax(axis = var_1497, x = qk_27_cast);
+            tensor<bool, []> var_1564_transpose_x_0 = const()[name = tensor<string, []>("op_1564_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1564_transpose_y_0 = const()[name = tensor<string, []>("op_1564_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_139 = transpose(perm = var_1558, x = var_1557_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1564_cast = matmul(transpose_x = var_1564_transpose_x_0, transpose_y = var_1564_transpose_y_0, x = var_1562_cast, y = transpose_139);
+            tensor<int32, [4]> var_1565 = const()[name = tensor<string, []>("op_1565"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_13 = const()[name = tensor<string, []>("concat_13"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_136 = transpose(perm = var_1565, x = var_1564_cast);
+            tensor<fp16, [1, 1500, 1024]> x_167_cast = reshape(shape = concat_13, x = transpose_136);
+            tensor<fp16, [1024, 1024]> var_1570_to_fp16 = const()[name = tensor<string, []>("op_1570_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343676800)))];
+            tensor<fp16, [1024]> var_1571_to_fp16 = const()[name = tensor<string, []>("op_1571_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345774016)))];
+            tensor<fp16, [1, 1500, 1024]> var_1572_cast = linear(bias = var_1571_to_fp16, weight = var_1570_to_fp16, x = x_167_cast);
+            tensor<fp16, [1, 1500, 1024]> x_169_cast = add(x = x_163_cast, y = var_1572_cast);
+            tensor<int32, [1]> var_1578_axes_0 = const()[name = tensor<string, []>("op_1578_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_13_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345776128)))];
+            tensor<fp16, [1024]> blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345778240)))];
+            tensor<fp16, [1, 1500, 1024]> var_1578_cast = layer_norm(axes = var_1578_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1503_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast);
+            tensor<fp16, [4096, 1024]> var_1587_to_fp16 = const()[name = tensor<string, []>("op_1587_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345780352)))];
+            tensor<fp16, [4096]> var_1588_to_fp16 = const()[name = tensor<string, []>("op_1588_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(354169024)))];
+            tensor<fp16, [1, 1500, 4096]> input_113_cast = linear(bias = var_1588_to_fp16, weight = var_1587_to_fp16, x = var_1578_cast);
+            tensor<string, []> x_173_mode_0 = const()[name = tensor<string, []>("x_173_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_173_cast = gelu(mode = x_173_mode_0, x = input_113_cast);
+            tensor<fp16, [1024, 4096]> var_1593_to_fp16 = const()[name = tensor<string, []>("op_1593_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(354177280)))];
+            tensor<fp16, [1024]> var_1594_to_fp16 = const()[name = tensor<string, []>("op_1594_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362565952)))];
+            tensor<fp16, [1, 1500, 1024]> var_1595_cast = linear(bias = var_1594_to_fp16, weight = var_1593_to_fp16, x = x_173_cast);
+            tensor<fp16, [1, 1500, 1024]> x_175_cast = add(x = x_169_cast, y = var_1595_cast);
+            tensor<int32, []> var_1604 = const()[name = tensor<string, []>("op_1604"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1621_axes_0 = const()[name = tensor<string, []>("op_1621_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_14_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362568064)))];
+            tensor<fp16, [1024]> blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362570176)))];
+            tensor<fp16, []> var_1610_to_fp16 = const()[name = tensor<string, []>("op_1610_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1621_cast = layer_norm(axes = var_1621_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1610_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast);
+            tensor<fp16, [1024, 1024]> var_1632_to_fp16 = const()[name = tensor<string, []>("op_1632_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362572288)))];
+            tensor<fp16, [1024]> var_1633_to_fp16 = const()[name = tensor<string, []>("op_1633_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(364669504)))];
+            tensor<fp16, [1, 1500, 1024]> q_57_cast = linear(bias = var_1633_to_fp16, weight = var_1632_to_fp16, x = var_1621_cast);
+            tensor<fp16, [1024, 1024]> var_1636_to_fp16 = const()[name = tensor<string, []>("op_1636_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(364671616)))];
+            tensor<fp16, [1024]> k_57_bias_0_to_fp16 = const()[name = tensor<string, []>("k_57_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(366768832)))];
+            tensor<fp16, [1, 1500, 1024]> k_57_cast = linear(bias = k_57_bias_0_to_fp16, weight = var_1636_to_fp16, x = var_1621_cast);
+            tensor<fp16, [1024, 1024]> var_1640_to_fp16 = const()[name = tensor<string, []>("op_1640_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(366770944)))];
+            tensor<fp16, [1024]> var_1641_to_fp16 = const()[name = tensor<string, []>("op_1641_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368868160)))];
+            tensor<fp16, [1, 1500, 1024]> v_57_cast = linear(bias = var_1641_to_fp16, weight = var_1640_to_fp16, x = var_1621_cast);
+            tensor<int32, [4]> var_1649 = const()[name = tensor<string, []>("op_1649"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1650_cast = reshape(shape = var_1649, x = q_57_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_196_to_fp16 = const()[name = tensor<string, []>("const_196_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_59_cast = mul(x = var_1650_cast, y = const_196_to_fp16);
+            tensor<int32, [4]> var_1656 = const()[name = tensor<string, []>("op_1656"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1657_cast = reshape(shape = var_1656, x = k_57_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_197_to_fp16 = const()[name = tensor<string, []>("const_197_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_59_cast = mul(x = var_1657_cast, y = const_197_to_fp16);
+            tensor<int32, [4]> var_1663 = const()[name = tensor<string, []>("op_1663"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1664_cast = reshape(shape = var_1663, x = v_57_cast);
+            tensor<int32, [4]> var_1665 = const()[name = tensor<string, []>("op_1665"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_29_transpose_x_0 = const()[name = tensor<string, []>("qk_29_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_29_transpose_y_0 = const()[name = tensor<string, []>("qk_29_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_76_perm_0 = const()[name = tensor<string, []>("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_77_perm_0 = const()[name = tensor<string, []>("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_133 = transpose(perm = transpose_77_perm_0, x = k_59_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_134 = transpose(perm = transpose_76_perm_0, x = q_59_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_29_cast = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_134, y = transpose_133);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1669_cast = softmax(axis = var_1604, x = qk_29_cast);
+            tensor<bool, []> var_1671_transpose_x_0 = const()[name = tensor<string, []>("op_1671_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1671_transpose_y_0 = const()[name = tensor<string, []>("op_1671_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_135 = transpose(perm = var_1665, x = var_1664_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1671_cast = matmul(transpose_x = var_1671_transpose_x_0, transpose_y = var_1671_transpose_y_0, x = var_1669_cast, y = transpose_135);
+            tensor<int32, [4]> var_1672 = const()[name = tensor<string, []>("op_1672"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_14 = const()[name = tensor<string, []>("concat_14"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_132 = transpose(perm = var_1672, x = var_1671_cast);
+            tensor<fp16, [1, 1500, 1024]> x_179_cast = reshape(shape = concat_14, x = transpose_132);
+            tensor<fp16, [1024, 1024]> var_1677_to_fp16 = const()[name = tensor<string, []>("op_1677_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368870272)))];
+            tensor<fp16, [1024]> var_1678_to_fp16 = const()[name = tensor<string, []>("op_1678_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370967488)))];
+            tensor<fp16, [1, 1500, 1024]> var_1679_cast = linear(bias = var_1678_to_fp16, weight = var_1677_to_fp16, x = x_179_cast);
+            tensor<fp16, [1, 1500, 1024]> x_181_cast = add(x = x_175_cast, y = var_1679_cast);
+            tensor<int32, [1]> var_1685_axes_0 = const()[name = tensor<string, []>("op_1685_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_14_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370969600)))];
+            tensor<fp16, [1024]> blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370971712)))];
+            tensor<fp16, [1, 1500, 1024]> var_1685_cast = layer_norm(axes = var_1685_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1610_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast);
+            tensor<fp16, [4096, 1024]> var_1694_to_fp16 = const()[name = tensor<string, []>("op_1694_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370973824)))];
+            tensor<fp16, [4096]> var_1695_to_fp16 = const()[name = tensor<string, []>("op_1695_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(379362496)))];
+            tensor<fp16, [1, 1500, 4096]> input_121_cast = linear(bias = var_1695_to_fp16, weight = var_1694_to_fp16, x = var_1685_cast);
+            tensor<string, []> x_185_mode_0 = const()[name = tensor<string, []>("x_185_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_185_cast = gelu(mode = x_185_mode_0, x = input_121_cast);
+            tensor<fp16, [1024, 4096]> var_1700_to_fp16 = const()[name = tensor<string, []>("op_1700_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(379370752)))];
+            tensor<fp16, [1024]> var_1701_to_fp16 = const()[name = tensor<string, []>("op_1701_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387759424)))];
+            tensor<fp16, [1, 1500, 1024]> var_1702_cast = linear(bias = var_1701_to_fp16, weight = var_1700_to_fp16, x = x_185_cast);
+            tensor<fp16, [1, 1500, 1024]> x_187_cast = add(x = x_181_cast, y = var_1702_cast);
+            tensor<int32, []> var_1711 = const()[name = tensor<string, []>("op_1711"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1728_axes_0 = const()[name = tensor<string, []>("op_1728_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_15_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387761536)))];
+            tensor<fp16, [1024]> blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387763648)))];
+            tensor<fp16, []> var_1717_to_fp16 = const()[name = tensor<string, []>("op_1717_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1728_cast = layer_norm(axes = var_1728_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1717_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast);
+            tensor<fp16, [1024, 1024]> var_1739_to_fp16 = const()[name = tensor<string, []>("op_1739_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387765760)))];
+            tensor<fp16, [1024]> var_1740_to_fp16 = const()[name = tensor<string, []>("op_1740_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389862976)))];
+            tensor<fp16, [1, 1500, 1024]> q_61_cast = linear(bias = var_1740_to_fp16, weight = var_1739_to_fp16, x = var_1728_cast);
+            tensor<fp16, [1024, 1024]> var_1743_to_fp16 = const()[name = tensor<string, []>("op_1743_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389865088)))];
+            tensor<fp16, [1024]> k_61_bias_0_to_fp16 = const()[name = tensor<string, []>("k_61_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(391962304)))];
+            tensor<fp16, [1, 1500, 1024]> k_61_cast = linear(bias = k_61_bias_0_to_fp16, weight = var_1743_to_fp16, x = var_1728_cast);
+            tensor<fp16, [1024, 1024]> var_1747_to_fp16 = const()[name = tensor<string, []>("op_1747_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(391964416)))];
+            tensor<fp16, [1024]> var_1748_to_fp16 = const()[name = tensor<string, []>("op_1748_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(394061632)))];
+            tensor<fp16, [1, 1500, 1024]> v_61_cast = linear(bias = var_1748_to_fp16, weight = var_1747_to_fp16, x = var_1728_cast);
+            tensor<int32, [4]> var_1756 = const()[name = tensor<string, []>("op_1756"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1757_cast = reshape(shape = var_1756, x = q_61_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_198_to_fp16 = const()[name = tensor<string, []>("const_198_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_63_cast = mul(x = var_1757_cast, y = const_198_to_fp16);
+            tensor<int32, [4]> var_1763 = const()[name = tensor<string, []>("op_1763"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1764_cast = reshape(shape = var_1763, x = k_61_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_199_to_fp16 = const()[name = tensor<string, []>("const_199_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_63_cast = mul(x = var_1764_cast, y = const_199_to_fp16);
+            tensor<int32, [4]> var_1770 = const()[name = tensor<string, []>("op_1770"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1771_cast = reshape(shape = var_1770, x = v_61_cast);
+            tensor<int32, [4]> var_1772 = const()[name = tensor<string, []>("op_1772"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_31_transpose_x_0 = const()[name = tensor<string, []>("qk_31_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_31_transpose_y_0 = const()[name = tensor<string, []>("qk_31_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_78_perm_0 = const()[name = tensor<string, []>("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_79_perm_0 = const()[name = tensor<string, []>("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_129 = transpose(perm = transpose_79_perm_0, x = k_63_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_130 = transpose(perm = transpose_78_perm_0, x = q_63_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_31_cast = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_130, y = transpose_129);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1776_cast = softmax(axis = var_1711, x = qk_31_cast);
+            tensor<bool, []> var_1778_transpose_x_0 = const()[name = tensor<string, []>("op_1778_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1778_transpose_y_0 = const()[name = tensor<string, []>("op_1778_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_131 = transpose(perm = var_1772, x = var_1771_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1778_cast = matmul(transpose_x = var_1778_transpose_x_0, transpose_y = var_1778_transpose_y_0, x = var_1776_cast, y = transpose_131);
+            tensor<int32, [4]> var_1779 = const()[name = tensor<string, []>("op_1779"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_15 = const()[name = tensor<string, []>("concat_15"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_128 = transpose(perm = var_1779, x = var_1778_cast);
+            tensor<fp16, [1, 1500, 1024]> x_191_cast = reshape(shape = concat_15, x = transpose_128);
+            tensor<fp16, [1024, 1024]> var_1784_to_fp16 = const()[name = tensor<string, []>("op_1784_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(394063744)))];
+            tensor<fp16, [1024]> var_1785_to_fp16 = const()[name = tensor<string, []>("op_1785_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396160960)))];
+            tensor<fp16, [1, 1500, 1024]> var_1786_cast = linear(bias = var_1785_to_fp16, weight = var_1784_to_fp16, x = x_191_cast);
+            tensor<fp16, [1, 1500, 1024]> x_193_cast = add(x = x_187_cast, y = var_1786_cast);
+            tensor<int32, [1]> var_1792_axes_0 = const()[name = tensor<string, []>("op_1792_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_15_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396163072)))];
+            tensor<fp16, [1024]> blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396165184)))];
+            tensor<fp16, [1, 1500, 1024]> var_1792_cast = layer_norm(axes = var_1792_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1717_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast);
+            tensor<fp16, [4096, 1024]> var_1801_to_fp16 = const()[name = tensor<string, []>("op_1801_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396167296)))];
+            tensor<fp16, [4096]> var_1802_to_fp16 = const()[name = tensor<string, []>("op_1802_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(404555968)))];
+            tensor<fp16, [1, 1500, 4096]> input_129_cast = linear(bias = var_1802_to_fp16, weight = var_1801_to_fp16, x = var_1792_cast);
+            tensor<string, []> x_197_mode_0 = const()[name = tensor<string, []>("x_197_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_197_cast = gelu(mode = x_197_mode_0, x = input_129_cast);
+            tensor<fp16, [1024, 4096]> var_1807_to_fp16 = const()[name = tensor<string, []>("op_1807_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(404564224)))];
+            tensor<fp16, [1024]> var_1808_to_fp16 = const()[name = tensor<string, []>("op_1808_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412952896)))];
+            tensor<fp16, [1, 1500, 1024]> var_1809_cast = linear(bias = var_1808_to_fp16, weight = var_1807_to_fp16, x = x_197_cast);
+            tensor<fp16, [1, 1500, 1024]> x_199_cast = add(x = x_193_cast, y = var_1809_cast);
+            tensor<int32, []> var_1818 = const()[name = tensor<string, []>("op_1818"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1835_axes_0 = const()[name = tensor<string, []>("op_1835_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_16_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412955008)))];
+            tensor<fp16, [1024]> blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412957120)))];
+            tensor<fp16, []> var_1824_to_fp16 = const()[name = tensor<string, []>("op_1824_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1835_cast = layer_norm(axes = var_1835_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1824_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast);
+            tensor<fp16, [1024, 1024]> var_1846_to_fp16 = const()[name = tensor<string, []>("op_1846_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412959232)))];
+            tensor<fp16, [1024]> var_1847_to_fp16 = const()[name = tensor<string, []>("op_1847_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415056448)))];
+            tensor<fp16, [1, 1500, 1024]> q_65_cast = linear(bias = var_1847_to_fp16, weight = var_1846_to_fp16, x = var_1835_cast);
+            tensor<fp16, [1024, 1024]> var_1850_to_fp16 = const()[name = tensor<string, []>("op_1850_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415058560)))];
+            tensor<fp16, [1024]> k_65_bias_0_to_fp16 = const()[name = tensor<string, []>("k_65_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417155776)))];
+            tensor<fp16, [1, 1500, 1024]> k_65_cast = linear(bias = k_65_bias_0_to_fp16, weight = var_1850_to_fp16, x = var_1835_cast);
+            tensor<fp16, [1024, 1024]> var_1854_to_fp16 = const()[name = tensor<string, []>("op_1854_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417157888)))];
+            tensor<fp16, [1024]> var_1855_to_fp16 = const()[name = tensor<string, []>("op_1855_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419255104)))];
+            tensor<fp16, [1, 1500, 1024]> v_65_cast = linear(bias = var_1855_to_fp16, weight = var_1854_to_fp16, x = var_1835_cast);
+            tensor<int32, [4]> var_1863 = const()[name = tensor<string, []>("op_1863"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1864_cast = reshape(shape = var_1863, x = q_65_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_200_to_fp16 = const()[name = tensor<string, []>("const_200_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_67_cast = mul(x = var_1864_cast, y = const_200_to_fp16);
+            tensor<int32, [4]> var_1870 = const()[name = tensor<string, []>("op_1870"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1871_cast = reshape(shape = var_1870, x = k_65_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_201_to_fp16 = const()[name = tensor<string, []>("const_201_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_67_cast = mul(x = var_1871_cast, y = const_201_to_fp16);
+            tensor<int32, [4]> var_1877 = const()[name = tensor<string, []>("op_1877"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1878_cast = reshape(shape = var_1877, x = v_65_cast);
+            tensor<int32, [4]> var_1879 = const()[name = tensor<string, []>("op_1879"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_33_transpose_x_0 = const()[name = tensor<string, []>("qk_33_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_33_transpose_y_0 = const()[name = tensor<string, []>("qk_33_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_80_perm_0 = const()[name = tensor<string, []>("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_125 = transpose(perm = transpose_81_perm_0, x = k_67_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_126 = transpose(perm = transpose_80_perm_0, x = q_67_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_33_cast = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_126, y = transpose_125);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1883_cast = softmax(axis = var_1818, x = qk_33_cast);
+            tensor<bool, []> var_1885_transpose_x_0 = const()[name = tensor<string, []>("op_1885_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1885_transpose_y_0 = const()[name = tensor<string, []>("op_1885_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_127 = transpose(perm = var_1879, x = var_1878_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1885_cast = matmul(transpose_x = var_1885_transpose_x_0, transpose_y = var_1885_transpose_y_0, x = var_1883_cast, y = transpose_127);
+            tensor<int32, [4]> var_1886 = const()[name = tensor<string, []>("op_1886"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_16 = const()[name = tensor<string, []>("concat_16"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_124 = transpose(perm = var_1886, x = var_1885_cast);
+            tensor<fp16, [1, 1500, 1024]> x_203_cast = reshape(shape = concat_16, x = transpose_124);
+            tensor<fp16, [1024, 1024]> var_1891_to_fp16 = const()[name = tensor<string, []>("op_1891_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419257216)))];
+            tensor<fp16, [1024]> var_1892_to_fp16 = const()[name = tensor<string, []>("op_1892_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421354432)))];
+            tensor<fp16, [1, 1500, 1024]> var_1893_cast = linear(bias = var_1892_to_fp16, weight = var_1891_to_fp16, x = x_203_cast);
+            tensor<fp16, [1, 1500, 1024]> x_205_cast = add(x = x_199_cast, y = var_1893_cast);
+            tensor<int32, [1]> var_1899_axes_0 = const()[name = tensor<string, []>("op_1899_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_16_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421356544)))];
+            tensor<fp16, [1024]> blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421358656)))];
+            tensor<fp16, [1, 1500, 1024]> var_1899_cast = layer_norm(axes = var_1899_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1824_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast);
+            tensor<fp16, [4096, 1024]> var_1908_to_fp16 = const()[name = tensor<string, []>("op_1908_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421360768)))];
+            tensor<fp16, [4096]> var_1909_to_fp16 = const()[name = tensor<string, []>("op_1909_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429749440)))];
+            tensor<fp16, [1, 1500, 4096]> input_137_cast = linear(bias = var_1909_to_fp16, weight = var_1908_to_fp16, x = var_1899_cast);
+            tensor<string, []> x_209_mode_0 = const()[name = tensor<string, []>("x_209_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_209_cast = gelu(mode = x_209_mode_0, x = input_137_cast);
+            tensor<fp16, [1024, 4096]> var_1914_to_fp16 = const()[name = tensor<string, []>("op_1914_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429757696)))];
+            tensor<fp16, [1024]> var_1915_to_fp16 = const()[name = tensor<string, []>("op_1915_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438146368)))];
+            tensor<fp16, [1, 1500, 1024]> var_1916_cast = linear(bias = var_1915_to_fp16, weight = var_1914_to_fp16, x = x_209_cast);
+            tensor<fp16, [1, 1500, 1024]> x_211_cast = add(x = x_205_cast, y = var_1916_cast);
+            tensor<int32, []> var_1925 = const()[name = tensor<string, []>("op_1925"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_1942_axes_0 = const()[name = tensor<string, []>("op_1942_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_17_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438148480)))];
+            tensor<fp16, [1024]> blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438150592)))];
+            tensor<fp16, []> var_1931_to_fp16 = const()[name = tensor<string, []>("op_1931_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_1942_cast = layer_norm(axes = var_1942_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1931_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast);
+            tensor<fp16, [1024, 1024]> var_1953_to_fp16 = const()[name = tensor<string, []>("op_1953_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(438152704)))];
+            tensor<fp16, [1024]> var_1954_to_fp16 = const()[name = tensor<string, []>("op_1954_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440249920)))];
+            tensor<fp16, [1, 1500, 1024]> q_69_cast = linear(bias = var_1954_to_fp16, weight = var_1953_to_fp16, x = var_1942_cast);
+            tensor<fp16, [1024, 1024]> var_1957_to_fp16 = const()[name = tensor<string, []>("op_1957_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440252032)))];
+            tensor<fp16, [1024]> k_69_bias_0_to_fp16 = const()[name = tensor<string, []>("k_69_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(442349248)))];
+            tensor<fp16, [1, 1500, 1024]> k_69_cast = linear(bias = k_69_bias_0_to_fp16, weight = var_1957_to_fp16, x = var_1942_cast);
+            tensor<fp16, [1024, 1024]> var_1961_to_fp16 = const()[name = tensor<string, []>("op_1961_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(442351360)))];
+            tensor<fp16, [1024]> var_1962_to_fp16 = const()[name = tensor<string, []>("op_1962_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444448576)))];
+            tensor<fp16, [1, 1500, 1024]> v_69_cast = linear(bias = var_1962_to_fp16, weight = var_1961_to_fp16, x = var_1942_cast);
+            tensor<int32, [4]> var_1970 = const()[name = tensor<string, []>("op_1970"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1971_cast = reshape(shape = var_1970, x = q_69_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_202_to_fp16 = const()[name = tensor<string, []>("const_202_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_71_cast = mul(x = var_1971_cast, y = const_202_to_fp16);
+            tensor<int32, [4]> var_1977 = const()[name = tensor<string, []>("op_1977"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1978_cast = reshape(shape = var_1977, x = k_69_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_203_to_fp16 = const()[name = tensor<string, []>("const_203_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_71_cast = mul(x = var_1978_cast, y = const_203_to_fp16);
+            tensor<int32, [4]> var_1984 = const()[name = tensor<string, []>("op_1984"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_1985_cast = reshape(shape = var_1984, x = v_69_cast);
+            tensor<int32, [4]> var_1986 = const()[name = tensor<string, []>("op_1986"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_35_transpose_x_0 = const()[name = tensor<string, []>("qk_35_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_35_transpose_y_0 = const()[name = tensor<string, []>("qk_35_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_121 = transpose(perm = transpose_83_perm_0, x = k_71_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_122 = transpose(perm = transpose_82_perm_0, x = q_71_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_35_cast = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_122, y = transpose_121);
+            tensor<fp16, [1, 16, 1500, 1500]> var_1990_cast = softmax(axis = var_1925, x = qk_35_cast);
+            tensor<bool, []> var_1992_transpose_x_0 = const()[name = tensor<string, []>("op_1992_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_1992_transpose_y_0 = const()[name = tensor<string, []>("op_1992_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_123 = transpose(perm = var_1986, x = var_1985_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_1992_cast = matmul(transpose_x = var_1992_transpose_x_0, transpose_y = var_1992_transpose_y_0, x = var_1990_cast, y = transpose_123);
+            tensor<int32, [4]> var_1993 = const()[name = tensor<string, []>("op_1993"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_17 = const()[name = tensor<string, []>("concat_17"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_120 = transpose(perm = var_1993, x = var_1992_cast);
+            tensor<fp16, [1, 1500, 1024]> x_215_cast = reshape(shape = concat_17, x = transpose_120);
+            tensor<fp16, [1024, 1024]> var_1998_to_fp16 = const()[name = tensor<string, []>("op_1998_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444450688)))];
+            tensor<fp16, [1024]> var_1999_to_fp16 = const()[name = tensor<string, []>("op_1999_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446547904)))];
+            tensor<fp16, [1, 1500, 1024]> var_2000_cast = linear(bias = var_1999_to_fp16, weight = var_1998_to_fp16, x = x_215_cast);
+            tensor<fp16, [1, 1500, 1024]> x_217_cast = add(x = x_211_cast, y = var_2000_cast);
+            tensor<int32, [1]> var_2006_axes_0 = const()[name = tensor<string, []>("op_2006_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_17_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446550016)))];
+            tensor<fp16, [1024]> blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446552128)))];
+            tensor<fp16, [1, 1500, 1024]> var_2006_cast = layer_norm(axes = var_2006_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1931_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast);
+            tensor<fp16, [4096, 1024]> var_2015_to_fp16 = const()[name = tensor<string, []>("op_2015_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(446554240)))];
+            tensor<fp16, [4096]> var_2016_to_fp16 = const()[name = tensor<string, []>("op_2016_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454942912)))];
+            tensor<fp16, [1, 1500, 4096]> input_145_cast = linear(bias = var_2016_to_fp16, weight = var_2015_to_fp16, x = var_2006_cast);
+            tensor<string, []> x_221_mode_0 = const()[name = tensor<string, []>("x_221_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_221_cast = gelu(mode = x_221_mode_0, x = input_145_cast);
+            tensor<fp16, [1024, 4096]> var_2021_to_fp16 = const()[name = tensor<string, []>("op_2021_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454951168)))];
+            tensor<fp16, [1024]> var_2022_to_fp16 = const()[name = tensor<string, []>("op_2022_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463339840)))];
+            tensor<fp16, [1, 1500, 1024]> var_2023_cast = linear(bias = var_2022_to_fp16, weight = var_2021_to_fp16, x = x_221_cast);
+            tensor<fp16, [1, 1500, 1024]> x_223_cast = add(x = x_217_cast, y = var_2023_cast);
+            tensor<int32, []> var_2032 = const()[name = tensor<string, []>("op_2032"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_2049_axes_0 = const()[name = tensor<string, []>("op_2049_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_18_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463341952)))];
+            tensor<fp16, [1024]> blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463344064)))];
+            tensor<fp16, []> var_2038_to_fp16 = const()[name = tensor<string, []>("op_2038_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_2049_cast = layer_norm(axes = var_2049_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast);
+            tensor<fp16, [1024, 1024]> var_2060_to_fp16 = const()[name = tensor<string, []>("op_2060_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(463346176)))];
+            tensor<fp16, [1024]> var_2061_to_fp16 = const()[name = tensor<string, []>("op_2061_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465443392)))];
+            tensor<fp16, [1, 1500, 1024]> q_73_cast = linear(bias = var_2061_to_fp16, weight = var_2060_to_fp16, x = var_2049_cast);
+            tensor<fp16, [1024, 1024]> var_2064_to_fp16 = const()[name = tensor<string, []>("op_2064_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465445504)))];
+            tensor<fp16, [1024]> k_73_bias_0_to_fp16 = const()[name = tensor<string, []>("k_73_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467542720)))];
+            tensor<fp16, [1, 1500, 1024]> k_73_cast = linear(bias = k_73_bias_0_to_fp16, weight = var_2064_to_fp16, x = var_2049_cast);
+            tensor<fp16, [1024, 1024]> var_2068_to_fp16 = const()[name = tensor<string, []>("op_2068_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467544832)))];
+            tensor<fp16, [1024]> var_2069_to_fp16 = const()[name = tensor<string, []>("op_2069_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(469642048)))];
+            tensor<fp16, [1, 1500, 1024]> v_73_cast = linear(bias = var_2069_to_fp16, weight = var_2068_to_fp16, x = var_2049_cast);
+            tensor<int32, [4]> var_2077 = const()[name = tensor<string, []>("op_2077"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2078_cast = reshape(shape = var_2077, x = q_73_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_204_to_fp16 = const()[name = tensor<string, []>("const_204_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_75_cast = mul(x = var_2078_cast, y = const_204_to_fp16);
+            tensor<int32, [4]> var_2084 = const()[name = tensor<string, []>("op_2084"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2085_cast = reshape(shape = var_2084, x = k_73_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_205_to_fp16 = const()[name = tensor<string, []>("const_205_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_75_cast = mul(x = var_2085_cast, y = const_205_to_fp16);
+            tensor<int32, [4]> var_2091 = const()[name = tensor<string, []>("op_2091"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2092_cast = reshape(shape = var_2091, x = v_73_cast);
+            tensor<int32, [4]> var_2093 = const()[name = tensor<string, []>("op_2093"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_37_transpose_x_0 = const()[name = tensor<string, []>("qk_37_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_37_transpose_y_0 = const()[name = tensor<string, []>("qk_37_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_117 = transpose(perm = transpose_85_perm_0, x = k_75_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_118 = transpose(perm = transpose_84_perm_0, x = q_75_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_37_cast = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_118, y = transpose_117);
+            tensor<fp16, [1, 16, 1500, 1500]> var_2097_cast = softmax(axis = var_2032, x = qk_37_cast);
+            tensor<bool, []> var_2099_transpose_x_0 = const()[name = tensor<string, []>("op_2099_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_2099_transpose_y_0 = const()[name = tensor<string, []>("op_2099_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_119 = transpose(perm = var_2093, x = var_2092_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_2099_cast = matmul(transpose_x = var_2099_transpose_x_0, transpose_y = var_2099_transpose_y_0, x = var_2097_cast, y = transpose_119);
+            tensor<int32, [4]> var_2100 = const()[name = tensor<string, []>("op_2100"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_18 = const()[name = tensor<string, []>("concat_18"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_116 = transpose(perm = var_2100, x = var_2099_cast);
+            tensor<fp16, [1, 1500, 1024]> x_227_cast = reshape(shape = concat_18, x = transpose_116);
+            tensor<fp16, [1024, 1024]> var_2105_to_fp16 = const()[name = tensor<string, []>("op_2105_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(469644160)))];
+            tensor<fp16, [1024]> var_2106_to_fp16 = const()[name = tensor<string, []>("op_2106_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471741376)))];
+            tensor<fp16, [1, 1500, 1024]> var_2107_cast = linear(bias = var_2106_to_fp16, weight = var_2105_to_fp16, x = x_227_cast);
+            tensor<fp16, [1, 1500, 1024]> x_229_cast = add(x = x_223_cast, y = var_2107_cast);
+            tensor<int32, [1]> var_2113_axes_0 = const()[name = tensor<string, []>("op_2113_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_18_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471743488)))];
+            tensor<fp16, [1024]> blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471745600)))];
+            tensor<fp16, [1, 1500, 1024]> var_2113_cast = layer_norm(axes = var_2113_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast);
+            tensor<fp16, [4096, 1024]> var_2122_to_fp16 = const()[name = tensor<string, []>("op_2122_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471747712)))];
+            tensor<fp16, [4096]> var_2123_to_fp16 = const()[name = tensor<string, []>("op_2123_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480136384)))];
+            tensor<fp16, [1, 1500, 4096]> input_153_cast = linear(bias = var_2123_to_fp16, weight = var_2122_to_fp16, x = var_2113_cast);
+            tensor<string, []> x_233_mode_0 = const()[name = tensor<string, []>("x_233_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_233_cast = gelu(mode = x_233_mode_0, x = input_153_cast);
+            tensor<fp16, [1024, 4096]> var_2128_to_fp16 = const()[name = tensor<string, []>("op_2128_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480144640)))];
+            tensor<fp16, [1024]> var_2129_to_fp16 = const()[name = tensor<string, []>("op_2129_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488533312)))];
+            tensor<fp16, [1, 1500, 1024]> var_2130_cast = linear(bias = var_2129_to_fp16, weight = var_2128_to_fp16, x = x_233_cast);
+            tensor<fp16, [1, 1500, 1024]> x_235_cast = add(x = x_229_cast, y = var_2130_cast);
+            tensor<int32, []> var_2139 = const()[name = tensor<string, []>("op_2139"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_2156_axes_0 = const()[name = tensor<string, []>("op_2156_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_19_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488535424)))];
+            tensor<fp16, [1024]> blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488537536)))];
+            tensor<fp16, []> var_2145_to_fp16 = const()[name = tensor<string, []>("op_2145_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_2156_cast = layer_norm(axes = var_2156_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2145_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast);
+            tensor<fp16, [1024, 1024]> var_2167_to_fp16 = const()[name = tensor<string, []>("op_2167_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488539648)))];
+            tensor<fp16, [1024]> var_2168_to_fp16 = const()[name = tensor<string, []>("op_2168_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490636864)))];
+            tensor<fp16, [1, 1500, 1024]> q_77_cast = linear(bias = var_2168_to_fp16, weight = var_2167_to_fp16, x = var_2156_cast);
+            tensor<fp16, [1024, 1024]> var_2171_to_fp16 = const()[name = tensor<string, []>("op_2171_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490638976)))];
+            tensor<fp16, [1024]> k_77_bias_0_to_fp16 = const()[name = tensor<string, []>("k_77_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492736192)))];
+            tensor<fp16, [1, 1500, 1024]> k_77_cast = linear(bias = k_77_bias_0_to_fp16, weight = var_2171_to_fp16, x = var_2156_cast);
+            tensor<fp16, [1024, 1024]> var_2175_to_fp16 = const()[name = tensor<string, []>("op_2175_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492738304)))];
+            tensor<fp16, [1024]> var_2176_to_fp16 = const()[name = tensor<string, []>("op_2176_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494835520)))];
+            tensor<fp16, [1, 1500, 1024]> v_77_cast = linear(bias = var_2176_to_fp16, weight = var_2175_to_fp16, x = var_2156_cast);
+            tensor<int32, [4]> var_2184 = const()[name = tensor<string, []>("op_2184"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2185_cast = reshape(shape = var_2184, x = q_77_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_206_to_fp16 = const()[name = tensor<string, []>("const_206_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_79_cast = mul(x = var_2185_cast, y = const_206_to_fp16);
+            tensor<int32, [4]> var_2191 = const()[name = tensor<string, []>("op_2191"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2192_cast = reshape(shape = var_2191, x = k_77_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_207_to_fp16 = const()[name = tensor<string, []>("const_207_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_79_cast = mul(x = var_2192_cast, y = const_207_to_fp16);
+            tensor<int32, [4]> var_2198 = const()[name = tensor<string, []>("op_2198"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2199_cast = reshape(shape = var_2198, x = v_77_cast);
+            tensor<int32, [4]> var_2200 = const()[name = tensor<string, []>("op_2200"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_39_transpose_x_0 = const()[name = tensor<string, []>("qk_39_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_39_transpose_y_0 = const()[name = tensor<string, []>("qk_39_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_87_perm_0 = const()[name = tensor<string, []>("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_113 = transpose(perm = transpose_87_perm_0, x = k_79_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_114 = transpose(perm = transpose_86_perm_0, x = q_79_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_39_cast = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_114, y = transpose_113);
+            tensor<fp16, [1, 16, 1500, 1500]> var_2204_cast = softmax(axis = var_2139, x = qk_39_cast);
+            tensor<bool, []> var_2206_transpose_x_0 = const()[name = tensor<string, []>("op_2206_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_2206_transpose_y_0 = const()[name = tensor<string, []>("op_2206_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_115 = transpose(perm = var_2200, x = var_2199_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_2206_cast = matmul(transpose_x = var_2206_transpose_x_0, transpose_y = var_2206_transpose_y_0, x = var_2204_cast, y = transpose_115);
+            tensor<int32, [4]> var_2207 = const()[name = tensor<string, []>("op_2207"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_19 = const()[name = tensor<string, []>("concat_19"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_112 = transpose(perm = var_2207, x = var_2206_cast);
+            tensor<fp16, [1, 1500, 1024]> x_239_cast = reshape(shape = concat_19, x = transpose_112);
+            tensor<fp16, [1024, 1024]> var_2212_to_fp16 = const()[name = tensor<string, []>("op_2212_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494837632)))];
+            tensor<fp16, [1024]> var_2213_to_fp16 = const()[name = tensor<string, []>("op_2213_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496934848)))];
+            tensor<fp16, [1, 1500, 1024]> var_2214_cast = linear(bias = var_2213_to_fp16, weight = var_2212_to_fp16, x = x_239_cast);
+            tensor<fp16, [1, 1500, 1024]> x_241_cast = add(x = x_235_cast, y = var_2214_cast);
+            tensor<int32, [1]> var_2220_axes_0 = const()[name = tensor<string, []>("op_2220_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_19_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496936960)))];
+            tensor<fp16, [1024]> blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496939072)))];
+            tensor<fp16, [1, 1500, 1024]> var_2220_cast = layer_norm(axes = var_2220_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2145_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast);
+            tensor<fp16, [4096, 1024]> var_2229_to_fp16 = const()[name = tensor<string, []>("op_2229_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496941184)))];
+            tensor<fp16, [4096]> var_2230_to_fp16 = const()[name = tensor<string, []>("op_2230_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505329856)))];
+            tensor<fp16, [1, 1500, 4096]> input_161_cast = linear(bias = var_2230_to_fp16, weight = var_2229_to_fp16, x = var_2220_cast);
+            tensor<string, []> x_245_mode_0 = const()[name = tensor<string, []>("x_245_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_245_cast = gelu(mode = x_245_mode_0, x = input_161_cast);
+            tensor<fp16, [1024, 4096]> var_2235_to_fp16 = const()[name = tensor<string, []>("op_2235_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505338112)))];
+            tensor<fp16, [1024]> var_2236_to_fp16 = const()[name = tensor<string, []>("op_2236_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513726784)))];
+            tensor<fp16, [1, 1500, 1024]> var_2237_cast = linear(bias = var_2236_to_fp16, weight = var_2235_to_fp16, x = x_245_cast);
+            tensor<fp16, [1, 1500, 1024]> x_247_cast = add(x = x_241_cast, y = var_2237_cast);
+            tensor<int32, []> var_2246 = const()[name = tensor<string, []>("op_2246"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_2263_axes_0 = const()[name = tensor<string, []>("op_2263_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_20_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513728896)))];
+            tensor<fp16, [1024]> blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513731008)))];
+            tensor<fp16, []> var_2252_to_fp16 = const()[name = tensor<string, []>("op_2252_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_2263_cast = layer_norm(axes = var_2263_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2252_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast);
+            tensor<fp16, [1024, 1024]> var_2274_to_fp16 = const()[name = tensor<string, []>("op_2274_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513733120)))];
+            tensor<fp16, [1024]> var_2275_to_fp16 = const()[name = tensor<string, []>("op_2275_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(515830336)))];
+            tensor<fp16, [1, 1500, 1024]> q_81_cast = linear(bias = var_2275_to_fp16, weight = var_2274_to_fp16, x = var_2263_cast);
+            tensor<fp16, [1024, 1024]> var_2278_to_fp16 = const()[name = tensor<string, []>("op_2278_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(515832448)))];
+            tensor<fp16, [1024]> k_81_bias_0_to_fp16 = const()[name = tensor<string, []>("k_81_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517929664)))];
+            tensor<fp16, [1, 1500, 1024]> k_81_cast = linear(bias = k_81_bias_0_to_fp16, weight = var_2278_to_fp16, x = var_2263_cast);
+            tensor<fp16, [1024, 1024]> var_2282_to_fp16 = const()[name = tensor<string, []>("op_2282_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517931776)))];
+            tensor<fp16, [1024]> var_2283_to_fp16 = const()[name = tensor<string, []>("op_2283_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520028992)))];
+            tensor<fp16, [1, 1500, 1024]> v_81_cast = linear(bias = var_2283_to_fp16, weight = var_2282_to_fp16, x = var_2263_cast);
+            tensor<int32, [4]> var_2291 = const()[name = tensor<string, []>("op_2291"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2292_cast = reshape(shape = var_2291, x = q_81_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_208_to_fp16 = const()[name = tensor<string, []>("const_208_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_83_cast = mul(x = var_2292_cast, y = const_208_to_fp16);
+            tensor<int32, [4]> var_2298 = const()[name = tensor<string, []>("op_2298"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2299_cast = reshape(shape = var_2298, x = k_81_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_209_to_fp16 = const()[name = tensor<string, []>("const_209_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_83_cast = mul(x = var_2299_cast, y = const_209_to_fp16);
+            tensor<int32, [4]> var_2305 = const()[name = tensor<string, []>("op_2305"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2306_cast = reshape(shape = var_2305, x = v_81_cast);
+            tensor<int32, [4]> var_2307 = const()[name = tensor<string, []>("op_2307"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_41_transpose_x_0 = const()[name = tensor<string, []>("qk_41_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_41_transpose_y_0 = const()[name = tensor<string, []>("qk_41_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_88_perm_0 = const()[name = tensor<string, []>("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_89_perm_0 = const()[name = tensor<string, []>("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_109 = transpose(perm = transpose_89_perm_0, x = k_83_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_110 = transpose(perm = transpose_88_perm_0, x = q_83_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_41_cast = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_110, y = transpose_109);
+            tensor<fp16, [1, 16, 1500, 1500]> var_2311_cast = softmax(axis = var_2246, x = qk_41_cast);
+            tensor<bool, []> var_2313_transpose_x_0 = const()[name = tensor<string, []>("op_2313_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_2313_transpose_y_0 = const()[name = tensor<string, []>("op_2313_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_111 = transpose(perm = var_2307, x = var_2306_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_2313_cast = matmul(transpose_x = var_2313_transpose_x_0, transpose_y = var_2313_transpose_y_0, x = var_2311_cast, y = transpose_111);
+            tensor<int32, [4]> var_2314 = const()[name = tensor<string, []>("op_2314"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_20 = const()[name = tensor<string, []>("concat_20"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_108 = transpose(perm = var_2314, x = var_2313_cast);
+            tensor<fp16, [1, 1500, 1024]> x_251_cast = reshape(shape = concat_20, x = transpose_108);
+            tensor<fp16, [1024, 1024]> var_2319_to_fp16 = const()[name = tensor<string, []>("op_2319_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520031104)))];
+            tensor<fp16, [1024]> var_2320_to_fp16 = const()[name = tensor<string, []>("op_2320_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522128320)))];
+            tensor<fp16, [1, 1500, 1024]> var_2321_cast = linear(bias = var_2320_to_fp16, weight = var_2319_to_fp16, x = x_251_cast);
+            tensor<fp16, [1, 1500, 1024]> x_253_cast = add(x = x_247_cast, y = var_2321_cast);
+            tensor<int32, [1]> var_2327_axes_0 = const()[name = tensor<string, []>("op_2327_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_20_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522130432)))];
+            tensor<fp16, [1024]> blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522132544)))];
+            tensor<fp16, [1, 1500, 1024]> var_2327_cast = layer_norm(axes = var_2327_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2252_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast);
+            tensor<fp16, [4096, 1024]> var_2336_to_fp16 = const()[name = tensor<string, []>("op_2336_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522134656)))];
+            tensor<fp16, [4096]> var_2337_to_fp16 = const()[name = tensor<string, []>("op_2337_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(530523328)))];
+            tensor<fp16, [1, 1500, 4096]> input_169_cast = linear(bias = var_2337_to_fp16, weight = var_2336_to_fp16, x = var_2327_cast);
+            tensor<string, []> x_257_mode_0 = const()[name = tensor<string, []>("x_257_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_257_cast = gelu(mode = x_257_mode_0, x = input_169_cast);
+            tensor<fp16, [1024, 4096]> var_2342_to_fp16 = const()[name = tensor<string, []>("op_2342_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(530531584)))];
+            tensor<fp16, [1024]> var_2343_to_fp16 = const()[name = tensor<string, []>("op_2343_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538920256)))];
+            tensor<fp16, [1, 1500, 1024]> var_2344_cast = linear(bias = var_2343_to_fp16, weight = var_2342_to_fp16, x = x_257_cast);
+            tensor<fp16, [1, 1500, 1024]> x_259_cast = add(x = x_253_cast, y = var_2344_cast);
+            tensor<int32, []> var_2353 = const()[name = tensor<string, []>("op_2353"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_2370_axes_0 = const()[name = tensor<string, []>("op_2370_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_21_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538922368)))];
+            tensor<fp16, [1024]> blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538924480)))];
+            tensor<fp16, []> var_2359_to_fp16 = const()[name = tensor<string, []>("op_2359_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_2370_cast = layer_norm(axes = var_2370_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2359_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast);
+            tensor<fp16, [1024, 1024]> var_2381_to_fp16 = const()[name = tensor<string, []>("op_2381_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538926592)))];
+            tensor<fp16, [1024]> var_2382_to_fp16 = const()[name = tensor<string, []>("op_2382_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541023808)))];
+            tensor<fp16, [1, 1500, 1024]> q_85_cast = linear(bias = var_2382_to_fp16, weight = var_2381_to_fp16, x = var_2370_cast);
+            tensor<fp16, [1024, 1024]> var_2385_to_fp16 = const()[name = tensor<string, []>("op_2385_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541025920)))];
+            tensor<fp16, [1024]> k_85_bias_0_to_fp16 = const()[name = tensor<string, []>("k_85_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543123136)))];
+            tensor<fp16, [1, 1500, 1024]> k_85_cast = linear(bias = k_85_bias_0_to_fp16, weight = var_2385_to_fp16, x = var_2370_cast);
+            tensor<fp16, [1024, 1024]> var_2389_to_fp16 = const()[name = tensor<string, []>("op_2389_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543125248)))];
+            tensor<fp16, [1024]> var_2390_to_fp16 = const()[name = tensor<string, []>("op_2390_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545222464)))];
+            tensor<fp16, [1, 1500, 1024]> v_85_cast = linear(bias = var_2390_to_fp16, weight = var_2389_to_fp16, x = var_2370_cast);
+            tensor<int32, [4]> var_2398 = const()[name = tensor<string, []>("op_2398"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2399_cast = reshape(shape = var_2398, x = q_85_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_210_to_fp16 = const()[name = tensor<string, []>("const_210_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_87_cast = mul(x = var_2399_cast, y = const_210_to_fp16);
+            tensor<int32, [4]> var_2405 = const()[name = tensor<string, []>("op_2405"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2406_cast = reshape(shape = var_2405, x = k_85_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_211_to_fp16 = const()[name = tensor<string, []>("const_211_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_87_cast = mul(x = var_2406_cast, y = const_211_to_fp16);
+            tensor<int32, [4]> var_2412 = const()[name = tensor<string, []>("op_2412"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2413_cast = reshape(shape = var_2412, x = v_85_cast);
+            tensor<int32, [4]> var_2414 = const()[name = tensor<string, []>("op_2414"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_43_transpose_x_0 = const()[name = tensor<string, []>("qk_43_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_43_transpose_y_0 = const()[name = tensor<string, []>("qk_43_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_90_perm_0 = const()[name = tensor<string, []>("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_91_perm_0 = const()[name = tensor<string, []>("transpose_91_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_105 = transpose(perm = transpose_91_perm_0, x = k_87_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_106 = transpose(perm = transpose_90_perm_0, x = q_87_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_43_cast = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_106, y = transpose_105);
+            tensor<fp16, [1, 16, 1500, 1500]> var_2418_cast = softmax(axis = var_2353, x = qk_43_cast);
+            tensor<bool, []> var_2420_transpose_x_0 = const()[name = tensor<string, []>("op_2420_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_2420_transpose_y_0 = const()[name = tensor<string, []>("op_2420_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_107 = transpose(perm = var_2414, x = var_2413_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_2420_cast = matmul(transpose_x = var_2420_transpose_x_0, transpose_y = var_2420_transpose_y_0, x = var_2418_cast, y = transpose_107);
+            tensor<int32, [4]> var_2421 = const()[name = tensor<string, []>("op_2421"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_21 = const()[name = tensor<string, []>("concat_21"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_104 = transpose(perm = var_2421, x = var_2420_cast);
+            tensor<fp16, [1, 1500, 1024]> x_263_cast = reshape(shape = concat_21, x = transpose_104);
+            tensor<fp16, [1024, 1024]> var_2426_to_fp16 = const()[name = tensor<string, []>("op_2426_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545224576)))];
+            tensor<fp16, [1024]> var_2427_to_fp16 = const()[name = tensor<string, []>("op_2427_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547321792)))];
+            tensor<fp16, [1, 1500, 1024]> var_2428_cast = linear(bias = var_2427_to_fp16, weight = var_2426_to_fp16, x = x_263_cast);
+            tensor<fp16, [1, 1500, 1024]> x_265_cast = add(x = x_259_cast, y = var_2428_cast);
+            tensor<int32, [1]> var_2434_axes_0 = const()[name = tensor<string, []>("op_2434_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_21_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547323904)))];
+            tensor<fp16, [1024]> blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547326016)))];
+            tensor<fp16, [1, 1500, 1024]> var_2434_cast = layer_norm(axes = var_2434_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2359_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast);
+            tensor<fp16, [4096, 1024]> var_2443_to_fp16 = const()[name = tensor<string, []>("op_2443_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(547328128)))];
+            tensor<fp16, [4096]> var_2444_to_fp16 = const()[name = tensor<string, []>("op_2444_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(555716800)))];
+            tensor<fp16, [1, 1500, 4096]> input_177_cast = linear(bias = var_2444_to_fp16, weight = var_2443_to_fp16, x = var_2434_cast);
+            tensor<string, []> x_269_mode_0 = const()[name = tensor<string, []>("x_269_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_269_cast = gelu(mode = x_269_mode_0, x = input_177_cast);
+            tensor<fp16, [1024, 4096]> var_2449_to_fp16 = const()[name = tensor<string, []>("op_2449_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(555725056)))];
+            tensor<fp16, [1024]> var_2450_to_fp16 = const()[name = tensor<string, []>("op_2450_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564113728)))];
+            tensor<fp16, [1, 1500, 1024]> var_2451_cast = linear(bias = var_2450_to_fp16, weight = var_2449_to_fp16, x = x_269_cast);
+            tensor<fp16, [1, 1500, 1024]> x_271_cast = add(x = x_265_cast, y = var_2451_cast);
+            tensor<int32, []> var_2460 = const()[name = tensor<string, []>("op_2460"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_2477_axes_0 = const()[name = tensor<string, []>("op_2477_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_22_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564115840)))];
+            tensor<fp16, [1024]> blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564117952)))];
+            tensor<fp16, []> var_2466_to_fp16 = const()[name = tensor<string, []>("op_2466_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_2477_cast = layer_norm(axes = var_2477_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2466_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast);
+            tensor<fp16, [1024, 1024]> var_2488_to_fp16 = const()[name = tensor<string, []>("op_2488_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564120064)))];
+            tensor<fp16, [1024]> var_2489_to_fp16 = const()[name = tensor<string, []>("op_2489_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566217280)))];
+            tensor<fp16, [1, 1500, 1024]> q_89_cast = linear(bias = var_2489_to_fp16, weight = var_2488_to_fp16, x = var_2477_cast);
+            tensor<fp16, [1024, 1024]> var_2492_to_fp16 = const()[name = tensor<string, []>("op_2492_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566219392)))];
+            tensor<fp16, [1024]> k_89_bias_0_to_fp16 = const()[name = tensor<string, []>("k_89_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568316608)))];
+            tensor<fp16, [1, 1500, 1024]> k_89_cast = linear(bias = k_89_bias_0_to_fp16, weight = var_2492_to_fp16, x = var_2477_cast);
+            tensor<fp16, [1024, 1024]> var_2496_to_fp16 = const()[name = tensor<string, []>("op_2496_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568318720)))];
+            tensor<fp16, [1024]> var_2497_to_fp16 = const()[name = tensor<string, []>("op_2497_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570415936)))];
+            tensor<fp16, [1, 1500, 1024]> v_89_cast = linear(bias = var_2497_to_fp16, weight = var_2496_to_fp16, x = var_2477_cast);
+            tensor<int32, [4]> var_2505 = const()[name = tensor<string, []>("op_2505"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2506_cast = reshape(shape = var_2505, x = q_89_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_212_to_fp16 = const()[name = tensor<string, []>("const_212_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_91_cast = mul(x = var_2506_cast, y = const_212_to_fp16);
+            tensor<int32, [4]> var_2512 = const()[name = tensor<string, []>("op_2512"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2513_cast = reshape(shape = var_2512, x = k_89_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_213_to_fp16 = const()[name = tensor<string, []>("const_213_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_91_cast = mul(x = var_2513_cast, y = const_213_to_fp16);
+            tensor<int32, [4]> var_2519 = const()[name = tensor<string, []>("op_2519"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2520_cast = reshape(shape = var_2519, x = v_89_cast);
+            tensor<int32, [4]> var_2521 = const()[name = tensor<string, []>("op_2521"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_45_transpose_x_0 = const()[name = tensor<string, []>("qk_45_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_45_transpose_y_0 = const()[name = tensor<string, []>("qk_45_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_92_perm_0 = const()[name = tensor<string, []>("transpose_92_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_93_perm_0 = const()[name = tensor<string, []>("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_101 = transpose(perm = transpose_93_perm_0, x = k_91_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_102 = transpose(perm = transpose_92_perm_0, x = q_91_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_45_cast = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_102, y = transpose_101);
+            tensor<fp16, [1, 16, 1500, 1500]> var_2525_cast = softmax(axis = var_2460, x = qk_45_cast);
+            tensor<bool, []> var_2527_transpose_x_0 = const()[name = tensor<string, []>("op_2527_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_2527_transpose_y_0 = const()[name = tensor<string, []>("op_2527_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_103 = transpose(perm = var_2521, x = var_2520_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_2527_cast = matmul(transpose_x = var_2527_transpose_x_0, transpose_y = var_2527_transpose_y_0, x = var_2525_cast, y = transpose_103);
+            tensor<int32, [4]> var_2528 = const()[name = tensor<string, []>("op_2528"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_22 = const()[name = tensor<string, []>("concat_22"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_100 = transpose(perm = var_2528, x = var_2527_cast);
+            tensor<fp16, [1, 1500, 1024]> x_275_cast = reshape(shape = concat_22, x = transpose_100);
+            tensor<fp16, [1024, 1024]> var_2533_to_fp16 = const()[name = tensor<string, []>("op_2533_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(570418048)))];
+            tensor<fp16, [1024]> var_2534_to_fp16 = const()[name = tensor<string, []>("op_2534_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572515264)))];
+            tensor<fp16, [1, 1500, 1024]> var_2535_cast = linear(bias = var_2534_to_fp16, weight = var_2533_to_fp16, x = x_275_cast);
+            tensor<fp16, [1, 1500, 1024]> x_277_cast = add(x = x_271_cast, y = var_2535_cast);
+            tensor<int32, [1]> var_2541_axes_0 = const()[name = tensor<string, []>("op_2541_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_22_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572517376)))];
+            tensor<fp16, [1024]> blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572519488)))];
+            tensor<fp16, [1, 1500, 1024]> var_2541_cast = layer_norm(axes = var_2541_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2466_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast);
+            tensor<fp16, [4096, 1024]> var_2550_to_fp16 = const()[name = tensor<string, []>("op_2550_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572521600)))];
+            tensor<fp16, [4096]> var_2551_to_fp16 = const()[name = tensor<string, []>("op_2551_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(580910272)))];
+            tensor<fp16, [1, 1500, 4096]> input_185_cast = linear(bias = var_2551_to_fp16, weight = var_2550_to_fp16, x = var_2541_cast);
+            tensor<string, []> x_281_mode_0 = const()[name = tensor<string, []>("x_281_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_281_cast = gelu(mode = x_281_mode_0, x = input_185_cast);
+            tensor<fp16, [1024, 4096]> var_2556_to_fp16 = const()[name = tensor<string, []>("op_2556_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(580918528)))];
+            tensor<fp16, [1024]> var_2557_to_fp16 = const()[name = tensor<string, []>("op_2557_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589307200)))];
+            tensor<fp16, [1, 1500, 1024]> var_2558_cast = linear(bias = var_2557_to_fp16, weight = var_2556_to_fp16, x = x_281_cast);
+            tensor<fp16, [1, 1500, 1024]> x_283_cast = add(x = x_277_cast, y = var_2558_cast);
+            tensor<int32, []> var_2567 = const()[name = tensor<string, []>("op_2567"), val = tensor<int32, []>(-1)];
+            tensor<int32, [1]> var_2584_axes_0 = const()[name = tensor<string, []>("op_2584_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_23_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589309312)))];
+            tensor<fp16, [1024]> blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589311424)))];
+            tensor<fp16, []> var_2573_to_fp16 = const()[name = tensor<string, []>("op_2573_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_2584_cast = layer_norm(axes = var_2584_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2573_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast);
+            tensor<fp16, [1024, 1024]> var_2595_to_fp16 = const()[name = tensor<string, []>("op_2595_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589313536)))];
+            tensor<fp16, [1024]> var_2596_to_fp16 = const()[name = tensor<string, []>("op_2596_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591410752)))];
+            tensor<fp16, [1, 1500, 1024]> q_93_cast = linear(bias = var_2596_to_fp16, weight = var_2595_to_fp16, x = var_2584_cast);
+            tensor<fp16, [1024, 1024]> var_2599_to_fp16 = const()[name = tensor<string, []>("op_2599_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591412864)))];
+            tensor<fp16, [1024]> k_93_bias_0_to_fp16 = const()[name = tensor<string, []>("k_93_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(593510080)))];
+            tensor<fp16, [1, 1500, 1024]> k_93_cast = linear(bias = k_93_bias_0_to_fp16, weight = var_2599_to_fp16, x = var_2584_cast);
+            tensor<fp16, [1024, 1024]> var_2603_to_fp16 = const()[name = tensor<string, []>("op_2603_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(593512192)))];
+            tensor<fp16, [1024]> var_2604_to_fp16 = const()[name = tensor<string, []>("op_2604_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595609408)))];
+            tensor<fp16, [1, 1500, 1024]> v_93_cast = linear(bias = var_2604_to_fp16, weight = var_2603_to_fp16, x = var_2584_cast);
+            tensor<int32, [4]> var_2612 = const()[name = tensor<string, []>("op_2612"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2613_cast = reshape(shape = var_2612, x = q_93_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_214_to_fp16 = const()[name = tensor<string, []>("const_214_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> q_cast = mul(x = var_2613_cast, y = const_214_to_fp16);
+            tensor<int32, [4]> var_2619 = const()[name = tensor<string, []>("op_2619"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2620_cast = reshape(shape = var_2619, x = k_93_cast);
+            tensor<fp16, [1, 1, 1, 1]> const_215_to_fp16 = const()[name = tensor<string, []>("const_215_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
+            tensor<fp16, [1, 1500, 16, 64]> k_cast = mul(x = var_2620_cast, y = const_215_to_fp16);
+            tensor<int32, [4]> var_2626 = const()[name = tensor<string, []>("op_2626"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
+            tensor<fp16, [1, 1500, 16, 64]> var_2627_cast = reshape(shape = var_2626, x = v_93_cast);
+            tensor<int32, [4]> var_2628 = const()[name = tensor<string, []>("op_2628"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> qk_transpose_x_0 = const()[name = tensor<string, []>("qk_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> qk_transpose_y_0 = const()[name = tensor<string, []>("qk_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_94_perm_0 = const()[name = tensor<string, []>("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [4]> transpose_95_perm_0 = const()[name = tensor<string, []>("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
+            tensor<fp16, [1, 16, 64, 1500]> transpose_97 = transpose(perm = transpose_95_perm_0, x = k_cast);
+            tensor<fp16, [1, 16, 1500, 64]> transpose_98 = transpose(perm = transpose_94_perm_0, x = q_cast);
+            tensor<fp16, [1, 16, 1500, 1500]> qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_98, y = transpose_97);
+            tensor<fp16, [1, 16, 1500, 1500]> var_2632_cast = softmax(axis = var_2567, x = qk_cast);
+            tensor<bool, []> var_2634_transpose_x_0 = const()[name = tensor<string, []>("op_2634_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> var_2634_transpose_y_0 = const()[name = tensor<string, []>("op_2634_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 16, 1500, 64]> transpose_99 = transpose(perm = var_2628, x = var_2627_cast);
+            tensor<fp16, [1, 16, 1500, 64]> var_2634_cast = matmul(transpose_x = var_2634_transpose_x_0, transpose_y = var_2634_transpose_y_0, x = var_2632_cast, y = transpose_99);
+            tensor<int32, [4]> var_2635 = const()[name = tensor<string, []>("op_2635"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> concat_23 = const()[name = tensor<string, []>("concat_23"), val = tensor<int32, [3]>([1, 1500, 1024])];
+            tensor<fp16, [1, 1500, 16, 64]> transpose_96 = transpose(perm = var_2635, x = var_2634_cast);
+            tensor<fp16, [1, 1500, 1024]> x_287_cast = reshape(shape = concat_23, x = transpose_96);
+            tensor<fp16, [1024, 1024]> var_2640_to_fp16 = const()[name = tensor<string, []>("op_2640_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(595611520)))];
+            tensor<fp16, [1024]> var_2641_to_fp16 = const()[name = tensor<string, []>("op_2641_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597708736)))];
+            tensor<fp16, [1, 1500, 1024]> var_2642_cast = linear(bias = var_2641_to_fp16, weight = var_2640_to_fp16, x = x_287_cast);
+            tensor<fp16, [1, 1500, 1024]> x_289_cast = add(x = x_283_cast, y = var_2642_cast);
+            tensor<int32, [1]> var_2648_axes_0 = const()[name = tensor<string, []>("op_2648_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_23_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597710848)))];
+            tensor<fp16, [1024]> blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597712960)))];
+            tensor<fp16, [1, 1500, 1024]> var_2648_cast = layer_norm(axes = var_2648_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2573_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast);
+            tensor<fp16, [4096, 1024]> var_2657_to_fp16 = const()[name = tensor<string, []>("op_2657_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597715072)))];
+            tensor<fp16, [4096]> var_2658_to_fp16 = const()[name = tensor<string, []>("op_2658_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606103744)))];
+            tensor<fp16, [1, 1500, 4096]> input_193_cast = linear(bias = var_2658_to_fp16, weight = var_2657_to_fp16, x = var_2648_cast);
+            tensor<string, []> x_293_mode_0 = const()[name = tensor<string, []>("x_293_mode_0"), val = tensor<string, []>("EXACT")];
+            tensor<fp16, [1, 1500, 4096]> x_293_cast = gelu(mode = x_293_mode_0, x = input_193_cast);
+            tensor<fp16, [1024, 4096]> var_2663_to_fp16 = const()[name = tensor<string, []>("op_2663_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606112000)))];
+            tensor<fp16, [1024]> var_2664_to_fp16 = const()[name = tensor<string, []>("op_2664_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614500672)))];
+            tensor<fp16, [1, 1500, 1024]> var_2665_cast = linear(bias = var_2664_to_fp16, weight = var_2663_to_fp16, x = x_293_cast);
+            tensor<fp16, [1, 1500, 1024]> x_cast = add(x = x_289_cast, y = var_2665_cast);
+            tensor<int32, [1]> var_2678_axes_0 = const()[name = tensor<string, []>("op_2678_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614502784)))];
+            tensor<fp16, [1024]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614504896)))];
+            tensor<fp16, []> var_2669_to_fp16 = const()[name = tensor<string, []>("op_2669_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 1500, 1024]> var_2678_cast = layer_norm(axes = var_2678_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_2669_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast);
+            tensor<string, []> var_2678_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_2678_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [1, 1500, 1024]> output = cast(dtype = var_2678_cast_to_fp32_dtype_0, x = var_2678_cast);
+        } -> (output);
+}
\ No newline at end of file