program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}, {"coremltools-component-torch", "1.11.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.1"}})]
{
    func main<ios15>(tensor<fp32, [1, 128, 3000]> logmel_data) {
            tensor<int32, []> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> var_80 = const()[name = tensor<string, []>("op_80"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_82 = const()[name = tensor<string, []>("op_82"), val = tensor<int32, [1]>([1])];
            tensor<string, []> var_84_pad_type_0 = const()[name = tensor<string, []>("op_84_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> var_84_pad_0 = const()[name = tensor<string, []>("op_84_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, [1280, 128, 3]> weight_3_to_fp16 = const()[name = tensor<string, []>("weight_3_to_fp16"), val = tensor<fp16, [1280, 128, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<fp16, [1280]> bias_3_to_fp16 = const()[name = tensor<string, []>("bias_3_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(983168)))];
            tensor<fp16, [1, 128, 3000]> cast_193 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data)[name = tensor<string, []>("cast_193")];
            tensor<fp16, [1, 1280, 3000]> var_84_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_82, groups = var_72, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_80, weight = weight_3_to_fp16, x = cast_193)[name = tensor<string, []>("op_84_cast_fp16")];
            tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1280, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_84_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")];
            tensor<int32, []> var_88 = const()[name = tensor<string, []>("op_88"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> var_97 = const()[name = tensor<string, []>("op_97"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1]> var_99 = const()[name = tensor<string, []>("op_99"), val = tensor<int32, [1]>([1])];
            tensor<string, []> var_101_pad_type_0 = const()[name = tensor<string, []>("op_101_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> var_101_pad_0 = const()[name = tensor<string, []>("op_101_pad_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1280, 1280, 3]> weight_7_to_fp16 = const()[name = tensor<string, []>("weight_7_to_fp16"), val = tensor<fp16, [1280, 1280, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(985792)))];
            tensor<fp16, [1280]> bias_7_to_fp16 = const()[name = tensor<string, []>("bias_7_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10816256)))];
            tensor<fp16, [1, 1280, 1500]> var_101_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_99, groups = var_88, pad = var_101_pad_0, pad_type = var_101_pad_type_0, strides = var_97, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("op_101_cast_fp16")];
            tensor<string, []> x_3_mode_0 = const()[name = tensor<string, []>("x_3_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1280, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_101_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")];
            tensor<int32, [3]> var_106 = const()[name = tensor<string, []>("op_106"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1500, 1280]> positional_embedding_to_fp16 = const()[name = tensor<string, []>("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10818880)))];
            tensor<fp16, [1, 1500, 1280]> transpose_320 = transpose(perm = var_106, x = x_3_cast_fp16)[name = tensor<string, []>("transpose_320")];
            tensor<fp16, [1, 1500, 1280]> var_109_cast_fp16 = add(x = transpose_320, y = positional_embedding_to_fp16)[name = tensor<string, []>("op_109_cast_fp16")];
            tensor<int32, []> var_122 = const()[name = tensor<string, []>("op_122"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_139_axes_0 = const()[name = tensor<string, []>("op_139_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14658944)))];
            tensor<fp16, [1280]> blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14661568)))];
            tensor<fp16, []> var_128_to_fp16 = const()[name = tensor<string, []>("op_128_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_139_cast_fp16 = layer_norm(axes = var_139_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_109_cast_fp16)[name = tensor<string, []>("op_139_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_150_to_fp16 = const()[name = tensor<string, []>("op_150_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14664192)))];
            tensor<fp16, [1280]> var_151_to_fp16 = const()[name = tensor<string, []>("op_151_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17941056)))];
            tensor<fp16, [1, 1500, 1280]> linear_0_cast_fp16 = linear(bias = var_151_to_fp16, weight = var_150_to_fp16, x = var_139_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_154_to_fp16 = const()[name = tensor<string, []>("op_154_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17943680)))];
            tensor<fp16, [1280]> linear_1_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21220544)))];
            tensor<fp16, [1, 1500, 1280]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_154_to_fp16, x = var_139_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_158_to_fp16 = const()[name = tensor<string, []>("op_158_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21223168)))];
            tensor<fp16, [1280]> var_159_to_fp16 = const()[name = tensor<string, []>("op_159_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24500032)))];
            tensor<fp16, [1, 1500, 1280]> linear_2_cast_fp16 = linear(bias = var_159_to_fp16, weight = var_158_to_fp16, x = var_139_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> var_167 = const()[name = tensor<string, []>("op_167"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_168_cast_fp16 = reshape(shape = var_167, x = linear_0_cast_fp16)[name = tensor<string, []>("op_168_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_224_to_fp16 = const()[name = tensor<string, []>("const_224_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_3_cast_fp16 = mul(x = var_168_cast_fp16, y = const_224_to_fp16)[name = tensor<string, []>("q_3_cast_fp16")];
            tensor<int32, [4]> var_174 = const()[name = tensor<string, []>("op_174"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_175_cast_fp16 = reshape(shape = var_174, x = linear_1_cast_fp16)[name = tensor<string, []>("op_175_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_225_to_fp16 = const()[name = tensor<string, []>("const_225_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_3_cast_fp16 = mul(x = var_175_cast_fp16, y = const_225_to_fp16)[name = tensor<string, []>("k_3_cast_fp16")];
            tensor<int32, [4]> var_181 = const()[name = tensor<string, []>("op_181"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_182_cast_fp16 = reshape(shape = var_181, x = linear_2_cast_fp16)[name = tensor<string, []>("op_182_cast_fp16")];
            tensor<int32, [4]> var_183 = const()[name = tensor<string, []>("op_183"), 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_128_perm_0 = const()[name = tensor<string, []>("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_129_perm_0 = const()[name = tensor<string, []>("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_317 = transpose(perm = transpose_129_perm_0, x = k_3_cast_fp16)[name = tensor<string, []>("transpose_317")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_318 = transpose(perm = transpose_128_perm_0, x = q_3_cast_fp16)[name = tensor<string, []>("transpose_318")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_318, y = transpose_317)[name = tensor<string, []>("qk_1_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_187_cast_fp16 = softmax(axis = var_122, x = qk_1_cast_fp16)[name = tensor<string, []>("op_187_cast_fp16")];
            tensor<bool, []> var_189_transpose_x_0 = const()[name = tensor<string, []>("op_189_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_189_transpose_y_0 = const()[name = tensor<string, []>("op_189_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_319 = transpose(perm = var_183, x = var_182_cast_fp16)[name = tensor<string, []>("transpose_319")];
            tensor<fp16, [1, 20, 1500, 64]> var_189_cast_fp16 = matmul(transpose_x = var_189_transpose_x_0, transpose_y = var_189_transpose_y_0, x = var_187_cast_fp16, y = transpose_319)[name = tensor<string, []>("op_189_cast_fp16")];
            tensor<int32, [4]> var_190 = const()[name = tensor<string, []>("op_190"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_316 = transpose(perm = var_190, x = var_189_cast_fp16)[name = tensor<string, []>("transpose_316")];
            tensor<fp16, [1, 1500, 1280]> x_11_cast_fp16 = reshape(shape = concat_0, x = transpose_316)[name = tensor<string, []>("x_11_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_195_to_fp16 = const()[name = tensor<string, []>("op_195_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24502656)))];
            tensor<fp16, [1280]> var_196_to_fp16 = const()[name = tensor<string, []>("op_196_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27779520)))];
            tensor<fp16, [1, 1500, 1280]> linear_3_cast_fp16 = linear(bias = var_196_to_fp16, weight = var_195_to_fp16, x = x_11_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_13_cast_fp16 = add(x = var_109_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")];
            tensor<int32, [1]> var_203_axes_0 = const()[name = tensor<string, []>("op_203_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27782144)))];
            tensor<fp16, [1280]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27784768)))];
            tensor<fp16, [1, 1500, 1280]> var_203_cast_fp16 = layer_norm(axes = var_203_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor<string, []>("op_203_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_212_to_fp16 = const()[name = tensor<string, []>("op_212_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27787392)))];
            tensor<fp16, [5120]> var_213_to_fp16 = const()[name = tensor<string, []>("op_213_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40894656)))];
            tensor<fp16, [1, 1500, 5120]> linear_4_cast_fp16 = linear(bias = var_213_to_fp16, weight = var_212_to_fp16, x = var_203_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<string, []> x_17_mode_0 = const()[name = tensor<string, []>("x_17_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("x_17_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_218_to_fp16 = const()[name = tensor<string, []>("op_218_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40904960)))];
            tensor<fp16, [1280]> var_219_to_fp16 = const()[name = tensor<string, []>("op_219_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54012224)))];
            tensor<fp16, [1, 1500, 1280]> linear_5_cast_fp16 = linear(bias = var_219_to_fp16, weight = var_218_to_fp16, x = x_17_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("x_19_cast_fp16")];
            tensor<int32, []> var_229 = const()[name = tensor<string, []>("op_229"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_246_axes_0 = const()[name = tensor<string, []>("op_246_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54014848)))];
            tensor<fp16, [1280]> blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54017472)))];
            tensor<fp16, []> var_235_to_fp16 = const()[name = tensor<string, []>("op_235_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_246_cast_fp16 = layer_norm(axes = var_246_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = tensor<string, []>("op_246_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_257_to_fp16 = const()[name = tensor<string, []>("op_257_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54020096)))];
            tensor<fp16, [1280]> var_258_to_fp16 = const()[name = tensor<string, []>("op_258_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57296960)))];
            tensor<fp16, [1, 1500, 1280]> linear_6_cast_fp16 = linear(bias = var_258_to_fp16, weight = var_257_to_fp16, x = var_246_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_261_to_fp16 = const()[name = tensor<string, []>("op_261_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57299584)))];
            tensor<fp16, [1, 1500, 1280]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_261_to_fp16, x = var_246_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_265_to_fp16 = const()[name = tensor<string, []>("op_265_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60576448)))];
            tensor<fp16, [1280]> var_266_to_fp16 = const()[name = tensor<string, []>("op_266_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63853312)))];
            tensor<fp16, [1, 1500, 1280]> linear_8_cast_fp16 = linear(bias = var_266_to_fp16, weight = var_265_to_fp16, x = var_246_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
            tensor<int32, [4]> var_274 = const()[name = tensor<string, []>("op_274"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_275_cast_fp16 = reshape(shape = var_274, x = linear_6_cast_fp16)[name = tensor<string, []>("op_275_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_226_to_fp16 = const()[name = tensor<string, []>("const_226_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_7_cast_fp16 = mul(x = var_275_cast_fp16, y = const_226_to_fp16)[name = tensor<string, []>("q_7_cast_fp16")];
            tensor<int32, [4]> var_281 = const()[name = tensor<string, []>("op_281"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_282_cast_fp16 = reshape(shape = var_281, x = linear_7_cast_fp16)[name = tensor<string, []>("op_282_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_227_to_fp16 = const()[name = tensor<string, []>("const_227_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_7_cast_fp16 = mul(x = var_282_cast_fp16, y = const_227_to_fp16)[name = tensor<string, []>("k_7_cast_fp16")];
            tensor<int32, [4]> var_288 = const()[name = tensor<string, []>("op_288"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_289_cast_fp16 = reshape(shape = var_288, x = linear_8_cast_fp16)[name = tensor<string, []>("op_289_cast_fp16")];
            tensor<int32, [4]> var_290 = const()[name = tensor<string, []>("op_290"), 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_130_perm_0 = const()[name = tensor<string, []>("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_131_perm_0 = const()[name = tensor<string, []>("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_313 = transpose(perm = transpose_131_perm_0, x = k_7_cast_fp16)[name = tensor<string, []>("transpose_313")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_314 = transpose(perm = transpose_130_perm_0, x = q_7_cast_fp16)[name = tensor<string, []>("transpose_314")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_314, y = transpose_313)[name = tensor<string, []>("qk_3_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_294_cast_fp16 = softmax(axis = var_229, x = qk_3_cast_fp16)[name = tensor<string, []>("op_294_cast_fp16")];
            tensor<bool, []> var_296_transpose_x_0 = const()[name = tensor<string, []>("op_296_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_296_transpose_y_0 = const()[name = tensor<string, []>("op_296_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_315 = transpose(perm = var_290, x = var_289_cast_fp16)[name = tensor<string, []>("transpose_315")];
            tensor<fp16, [1, 20, 1500, 64]> var_296_cast_fp16 = matmul(transpose_x = var_296_transpose_x_0, transpose_y = var_296_transpose_y_0, x = var_294_cast_fp16, y = transpose_315)[name = tensor<string, []>("op_296_cast_fp16")];
            tensor<int32, [4]> var_297 = const()[name = tensor<string, []>("op_297"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_312 = transpose(perm = var_297, x = var_296_cast_fp16)[name = tensor<string, []>("transpose_312")];
            tensor<fp16, [1, 1500, 1280]> x_23_cast_fp16 = reshape(shape = concat_1, x = transpose_312)[name = tensor<string, []>("x_23_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_302_to_fp16 = const()[name = tensor<string, []>("op_302_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63855936)))];
            tensor<fp16, [1280]> var_303_to_fp16 = const()[name = tensor<string, []>("op_303_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67132800)))];
            tensor<fp16, [1, 1500, 1280]> linear_9_cast_fp16 = linear(bias = var_303_to_fp16, weight = var_302_to_fp16, x = x_23_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("x_25_cast_fp16")];
            tensor<int32, [1]> var_310_axes_0 = const()[name = tensor<string, []>("op_310_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67135424)))];
            tensor<fp16, [1280]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67138048)))];
            tensor<fp16, [1, 1500, 1280]> var_310_cast_fp16 = layer_norm(axes = var_310_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = tensor<string, []>("op_310_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_319_to_fp16 = const()[name = tensor<string, []>("op_319_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67140672)))];
            tensor<fp16, [5120]> var_320_to_fp16 = const()[name = tensor<string, []>("op_320_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80247936)))];
            tensor<fp16, [1, 1500, 5120]> linear_10_cast_fp16 = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_310_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
            tensor<string, []> x_29_mode_0 = const()[name = tensor<string, []>("x_29_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("x_29_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_325_to_fp16 = const()[name = tensor<string, []>("op_325_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80258240)))];
            tensor<fp16, [1280]> var_326_to_fp16 = const()[name = tensor<string, []>("op_326_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93365504)))];
            tensor<fp16, [1, 1500, 1280]> linear_11_cast_fp16 = linear(bias = var_326_to_fp16, weight = var_325_to_fp16, x = x_29_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("x_31_cast_fp16")];
            tensor<int32, []> var_336 = const()[name = tensor<string, []>("op_336"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_353_axes_0 = const()[name = tensor<string, []>("op_353_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93368128)))];
            tensor<fp16, [1280]> blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93370752)))];
            tensor<fp16, []> var_342_to_fp16 = const()[name = tensor<string, []>("op_342_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_353_cast_fp16 = layer_norm(axes = var_353_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = tensor<string, []>("op_353_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_364_to_fp16 = const()[name = tensor<string, []>("op_364_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93373376)))];
            tensor<fp16, [1280]> var_365_to_fp16 = const()[name = tensor<string, []>("op_365_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96650240)))];
            tensor<fp16, [1, 1500, 1280]> linear_12_cast_fp16 = linear(bias = var_365_to_fp16, weight = var_364_to_fp16, x = var_353_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_368_to_fp16 = const()[name = tensor<string, []>("op_368_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96652864)))];
            tensor<fp16, [1, 1500, 1280]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_368_to_fp16, x = var_353_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_372_to_fp16 = const()[name = tensor<string, []>("op_372_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99929728)))];
            tensor<fp16, [1280]> var_373_to_fp16 = const()[name = tensor<string, []>("op_373_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103206592)))];
            tensor<fp16, [1, 1500, 1280]> linear_14_cast_fp16 = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = var_353_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<int32, [4]> var_381 = const()[name = tensor<string, []>("op_381"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_382_cast_fp16 = reshape(shape = var_381, x = linear_12_cast_fp16)[name = tensor<string, []>("op_382_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_228_to_fp16 = const()[name = tensor<string, []>("const_228_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_11_cast_fp16 = mul(x = var_382_cast_fp16, y = const_228_to_fp16)[name = tensor<string, []>("q_11_cast_fp16")];
            tensor<int32, [4]> var_388 = const()[name = tensor<string, []>("op_388"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_389_cast_fp16 = reshape(shape = var_388, x = linear_13_cast_fp16)[name = tensor<string, []>("op_389_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_229_to_fp16 = const()[name = tensor<string, []>("const_229_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_11_cast_fp16 = mul(x = var_389_cast_fp16, y = const_229_to_fp16)[name = tensor<string, []>("k_11_cast_fp16")];
            tensor<int32, [4]> var_395 = const()[name = tensor<string, []>("op_395"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_396_cast_fp16 = reshape(shape = var_395, x = linear_14_cast_fp16)[name = tensor<string, []>("op_396_cast_fp16")];
            tensor<int32, [4]> var_397 = const()[name = tensor<string, []>("op_397"), 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_132_perm_0 = const()[name = tensor<string, []>("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_133_perm_0 = const()[name = tensor<string, []>("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_309 = transpose(perm = transpose_133_perm_0, x = k_11_cast_fp16)[name = tensor<string, []>("transpose_309")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_310 = transpose(perm = transpose_132_perm_0, x = q_11_cast_fp16)[name = tensor<string, []>("transpose_310")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_310, y = transpose_309)[name = tensor<string, []>("qk_5_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_401_cast_fp16 = softmax(axis = var_336, x = qk_5_cast_fp16)[name = tensor<string, []>("op_401_cast_fp16")];
            tensor<bool, []> var_403_transpose_x_0 = const()[name = tensor<string, []>("op_403_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_403_transpose_y_0 = const()[name = tensor<string, []>("op_403_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_311 = transpose(perm = var_397, x = var_396_cast_fp16)[name = tensor<string, []>("transpose_311")];
            tensor<fp16, [1, 20, 1500, 64]> var_403_cast_fp16 = matmul(transpose_x = var_403_transpose_x_0, transpose_y = var_403_transpose_y_0, x = var_401_cast_fp16, y = transpose_311)[name = tensor<string, []>("op_403_cast_fp16")];
            tensor<int32, [4]> var_404 = const()[name = tensor<string, []>("op_404"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_308 = transpose(perm = var_404, x = var_403_cast_fp16)[name = tensor<string, []>("transpose_308")];
            tensor<fp16, [1, 1500, 1280]> x_35_cast_fp16 = reshape(shape = concat_2, x = transpose_308)[name = tensor<string, []>("x_35_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_409_to_fp16 = const()[name = tensor<string, []>("op_409_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103209216)))];
            tensor<fp16, [1280]> var_410_to_fp16 = const()[name = tensor<string, []>("op_410_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106486080)))];
            tensor<fp16, [1, 1500, 1280]> linear_15_cast_fp16 = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = x_35_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("x_37_cast_fp16")];
            tensor<int32, [1]> var_417_axes_0 = const()[name = tensor<string, []>("op_417_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106488704)))];
            tensor<fp16, [1280]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106491328)))];
            tensor<fp16, [1, 1500, 1280]> var_417_cast_fp16 = layer_norm(axes = var_417_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = tensor<string, []>("op_417_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_426_to_fp16 = const()[name = tensor<string, []>("op_426_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106493952)))];
            tensor<fp16, [5120]> var_427_to_fp16 = const()[name = tensor<string, []>("op_427_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119601216)))];
            tensor<fp16, [1, 1500, 5120]> linear_16_cast_fp16 = linear(bias = var_427_to_fp16, weight = var_426_to_fp16, x = var_417_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
            tensor<string, []> x_41_mode_0 = const()[name = tensor<string, []>("x_41_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = tensor<string, []>("x_41_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_432_to_fp16 = const()[name = tensor<string, []>("op_432_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119611520)))];
            tensor<fp16, [1280]> var_433_to_fp16 = const()[name = tensor<string, []>("op_433_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132718784)))];
            tensor<fp16, [1, 1500, 1280]> linear_17_cast_fp16 = linear(bias = var_433_to_fp16, weight = var_432_to_fp16, x = x_41_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("x_43_cast_fp16")];
            tensor<int32, []> var_443 = const()[name = tensor<string, []>("op_443"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_460_axes_0 = const()[name = tensor<string, []>("op_460_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132721408)))];
            tensor<fp16, [1280]> blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132724032)))];
            tensor<fp16, []> var_449_to_fp16 = const()[name = tensor<string, []>("op_449_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_460_cast_fp16 = layer_norm(axes = var_460_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = tensor<string, []>("op_460_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_471_to_fp16 = const()[name = tensor<string, []>("op_471_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132726656)))];
            tensor<fp16, [1280]> var_472_to_fp16 = const()[name = tensor<string, []>("op_472_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136003520)))];
            tensor<fp16, [1, 1500, 1280]> linear_18_cast_fp16 = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_460_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_475_to_fp16 = const()[name = tensor<string, []>("op_475_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136006144)))];
            tensor<fp16, [1, 1500, 1280]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_475_to_fp16, x = var_460_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_479_to_fp16 = const()[name = tensor<string, []>("op_479_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139283008)))];
            tensor<fp16, [1280]> var_480_to_fp16 = const()[name = tensor<string, []>("op_480_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142559872)))];
            tensor<fp16, [1, 1500, 1280]> linear_20_cast_fp16 = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = var_460_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<int32, [4]> var_488 = const()[name = tensor<string, []>("op_488"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_489_cast_fp16 = reshape(shape = var_488, x = linear_18_cast_fp16)[name = tensor<string, []>("op_489_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_230_to_fp16 = const()[name = tensor<string, []>("const_230_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_15_cast_fp16 = mul(x = var_489_cast_fp16, y = const_230_to_fp16)[name = tensor<string, []>("q_15_cast_fp16")];
            tensor<int32, [4]> var_495 = const()[name = tensor<string, []>("op_495"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_496_cast_fp16 = reshape(shape = var_495, x = linear_19_cast_fp16)[name = tensor<string, []>("op_496_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_231_to_fp16 = const()[name = tensor<string, []>("const_231_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_15_cast_fp16 = mul(x = var_496_cast_fp16, y = const_231_to_fp16)[name = tensor<string, []>("k_15_cast_fp16")];
            tensor<int32, [4]> var_502 = const()[name = tensor<string, []>("op_502"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_503_cast_fp16 = reshape(shape = var_502, x = linear_20_cast_fp16)[name = tensor<string, []>("op_503_cast_fp16")];
            tensor<int32, [4]> var_504 = const()[name = tensor<string, []>("op_504"), 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_134_perm_0 = const()[name = tensor<string, []>("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_135_perm_0 = const()[name = tensor<string, []>("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_305 = transpose(perm = transpose_135_perm_0, x = k_15_cast_fp16)[name = tensor<string, []>("transpose_305")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_306 = transpose(perm = transpose_134_perm_0, x = q_15_cast_fp16)[name = tensor<string, []>("transpose_306")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_306, y = transpose_305)[name = tensor<string, []>("qk_7_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_508_cast_fp16 = softmax(axis = var_443, x = qk_7_cast_fp16)[name = tensor<string, []>("op_508_cast_fp16")];
            tensor<bool, []> var_510_transpose_x_0 = const()[name = tensor<string, []>("op_510_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_510_transpose_y_0 = const()[name = tensor<string, []>("op_510_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_307 = transpose(perm = var_504, x = var_503_cast_fp16)[name = tensor<string, []>("transpose_307")];
            tensor<fp16, [1, 20, 1500, 64]> var_510_cast_fp16 = matmul(transpose_x = var_510_transpose_x_0, transpose_y = var_510_transpose_y_0, x = var_508_cast_fp16, y = transpose_307)[name = tensor<string, []>("op_510_cast_fp16")];
            tensor<int32, [4]> var_511 = const()[name = tensor<string, []>("op_511"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_304 = transpose(perm = var_511, x = var_510_cast_fp16)[name = tensor<string, []>("transpose_304")];
            tensor<fp16, [1, 1500, 1280]> x_47_cast_fp16 = reshape(shape = concat_3, x = transpose_304)[name = tensor<string, []>("x_47_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_516_to_fp16 = const()[name = tensor<string, []>("op_516_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142562496)))];
            tensor<fp16, [1280]> var_517_to_fp16 = const()[name = tensor<string, []>("op_517_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145839360)))];
            tensor<fp16, [1, 1500, 1280]> linear_21_cast_fp16 = linear(bias = var_517_to_fp16, weight = var_516_to_fp16, x = x_47_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("x_49_cast_fp16")];
            tensor<int32, [1]> var_524_axes_0 = const()[name = tensor<string, []>("op_524_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145841984)))];
            tensor<fp16, [1280]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145844608)))];
            tensor<fp16, [1, 1500, 1280]> var_524_cast_fp16 = layer_norm(axes = var_524_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = tensor<string, []>("op_524_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_533_to_fp16 = const()[name = tensor<string, []>("op_533_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145847232)))];
            tensor<fp16, [5120]> var_534_to_fp16 = const()[name = tensor<string, []>("op_534_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158954496)))];
            tensor<fp16, [1, 1500, 5120]> linear_22_cast_fp16 = linear(bias = var_534_to_fp16, weight = var_533_to_fp16, x = var_524_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
            tensor<string, []> x_53_mode_0 = const()[name = tensor<string, []>("x_53_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = tensor<string, []>("x_53_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_539_to_fp16 = const()[name = tensor<string, []>("op_539_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158964800)))];
            tensor<fp16, [1280]> var_540_to_fp16 = const()[name = tensor<string, []>("op_540_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172072064)))];
            tensor<fp16, [1, 1500, 1280]> linear_23_cast_fp16 = linear(bias = var_540_to_fp16, weight = var_539_to_fp16, x = x_53_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("x_55_cast_fp16")];
            tensor<int32, []> var_550 = const()[name = tensor<string, []>("op_550"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_567_axes_0 = const()[name = tensor<string, []>("op_567_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172074688)))];
            tensor<fp16, [1280]> blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172077312)))];
            tensor<fp16, []> var_556_to_fp16 = const()[name = tensor<string, []>("op_556_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_567_cast_fp16 = layer_norm(axes = var_567_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = tensor<string, []>("op_567_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_578_to_fp16 = const()[name = tensor<string, []>("op_578_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172079936)))];
            tensor<fp16, [1280]> var_579_to_fp16 = const()[name = tensor<string, []>("op_579_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175356800)))];
            tensor<fp16, [1, 1500, 1280]> linear_24_cast_fp16 = linear(bias = var_579_to_fp16, weight = var_578_to_fp16, x = var_567_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_582_to_fp16 = const()[name = tensor<string, []>("op_582_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175359424)))];
            tensor<fp16, [1, 1500, 1280]> linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_582_to_fp16, x = var_567_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_586_to_fp16 = const()[name = tensor<string, []>("op_586_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178636288)))];
            tensor<fp16, [1280]> var_587_to_fp16 = const()[name = tensor<string, []>("op_587_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181913152)))];
            tensor<fp16, [1, 1500, 1280]> linear_26_cast_fp16 = linear(bias = var_587_to_fp16, weight = var_586_to_fp16, x = var_567_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<int32, [4]> var_595 = const()[name = tensor<string, []>("op_595"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_596_cast_fp16 = reshape(shape = var_595, x = linear_24_cast_fp16)[name = tensor<string, []>("op_596_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_232_to_fp16 = const()[name = tensor<string, []>("const_232_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_19_cast_fp16 = mul(x = var_596_cast_fp16, y = const_232_to_fp16)[name = tensor<string, []>("q_19_cast_fp16")];
            tensor<int32, [4]> var_602 = const()[name = tensor<string, []>("op_602"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_603_cast_fp16 = reshape(shape = var_602, x = linear_25_cast_fp16)[name = tensor<string, []>("op_603_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_233_to_fp16 = const()[name = tensor<string, []>("const_233_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_19_cast_fp16 = mul(x = var_603_cast_fp16, y = const_233_to_fp16)[name = tensor<string, []>("k_19_cast_fp16")];
            tensor<int32, [4]> var_609 = const()[name = tensor<string, []>("op_609"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_610_cast_fp16 = reshape(shape = var_609, x = linear_26_cast_fp16)[name = tensor<string, []>("op_610_cast_fp16")];
            tensor<int32, [4]> var_611 = const()[name = tensor<string, []>("op_611"), 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_136_perm_0 = const()[name = tensor<string, []>("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_137_perm_0 = const()[name = tensor<string, []>("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_301 = transpose(perm = transpose_137_perm_0, x = k_19_cast_fp16)[name = tensor<string, []>("transpose_301")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_302 = transpose(perm = transpose_136_perm_0, x = q_19_cast_fp16)[name = tensor<string, []>("transpose_302")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_302, y = transpose_301)[name = tensor<string, []>("qk_9_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_615_cast_fp16 = softmax(axis = var_550, x = qk_9_cast_fp16)[name = tensor<string, []>("op_615_cast_fp16")];
            tensor<bool, []> var_617_transpose_x_0 = const()[name = tensor<string, []>("op_617_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_617_transpose_y_0 = const()[name = tensor<string, []>("op_617_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_303 = transpose(perm = var_611, x = var_610_cast_fp16)[name = tensor<string, []>("transpose_303")];
            tensor<fp16, [1, 20, 1500, 64]> var_617_cast_fp16 = matmul(transpose_x = var_617_transpose_x_0, transpose_y = var_617_transpose_y_0, x = var_615_cast_fp16, y = transpose_303)[name = tensor<string, []>("op_617_cast_fp16")];
            tensor<int32, [4]> var_618 = const()[name = tensor<string, []>("op_618"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_300 = transpose(perm = var_618, x = var_617_cast_fp16)[name = tensor<string, []>("transpose_300")];
            tensor<fp16, [1, 1500, 1280]> x_59_cast_fp16 = reshape(shape = concat_4, x = transpose_300)[name = tensor<string, []>("x_59_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_623_to_fp16 = const()[name = tensor<string, []>("op_623_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181915776)))];
            tensor<fp16, [1280]> var_624_to_fp16 = const()[name = tensor<string, []>("op_624_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185192640)))];
            tensor<fp16, [1, 1500, 1280]> linear_27_cast_fp16 = linear(bias = var_624_to_fp16, weight = var_623_to_fp16, x = x_59_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("x_61_cast_fp16")];
            tensor<int32, [1]> var_631_axes_0 = const()[name = tensor<string, []>("op_631_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185195264)))];
            tensor<fp16, [1280]> blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185197888)))];
            tensor<fp16, [1, 1500, 1280]> var_631_cast_fp16 = layer_norm(axes = var_631_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = tensor<string, []>("op_631_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_640_to_fp16 = const()[name = tensor<string, []>("op_640_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185200512)))];
            tensor<fp16, [5120]> var_641_to_fp16 = const()[name = tensor<string, []>("op_641_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198307776)))];
            tensor<fp16, [1, 1500, 5120]> linear_28_cast_fp16 = linear(bias = var_641_to_fp16, weight = var_640_to_fp16, x = var_631_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
            tensor<string, []> x_65_mode_0 = const()[name = tensor<string, []>("x_65_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = tensor<string, []>("x_65_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_646_to_fp16 = const()[name = tensor<string, []>("op_646_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198318080)))];
            tensor<fp16, [1280]> var_647_to_fp16 = const()[name = tensor<string, []>("op_647_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211425344)))];
            tensor<fp16, [1, 1500, 1280]> linear_29_cast_fp16 = linear(bias = var_647_to_fp16, weight = var_646_to_fp16, x = x_65_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("x_67_cast_fp16")];
            tensor<int32, []> var_657 = const()[name = tensor<string, []>("op_657"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_674_axes_0 = const()[name = tensor<string, []>("op_674_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211427968)))];
            tensor<fp16, [1280]> blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211430592)))];
            tensor<fp16, []> var_663_to_fp16 = const()[name = tensor<string, []>("op_663_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_674_cast_fp16 = layer_norm(axes = var_674_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = tensor<string, []>("op_674_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_685_to_fp16 = const()[name = tensor<string, []>("op_685_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211433216)))];
            tensor<fp16, [1280]> var_686_to_fp16 = const()[name = tensor<string, []>("op_686_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214710080)))];
            tensor<fp16, [1, 1500, 1280]> linear_30_cast_fp16 = linear(bias = var_686_to_fp16, weight = var_685_to_fp16, x = var_674_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_689_to_fp16 = const()[name = tensor<string, []>("op_689_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214712704)))];
            tensor<fp16, [1, 1500, 1280]> linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_689_to_fp16, x = var_674_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_693_to_fp16 = const()[name = tensor<string, []>("op_693_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217989568)))];
            tensor<fp16, [1280]> var_694_to_fp16 = const()[name = tensor<string, []>("op_694_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221266432)))];
            tensor<fp16, [1, 1500, 1280]> linear_32_cast_fp16 = linear(bias = var_694_to_fp16, weight = var_693_to_fp16, x = var_674_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<int32, [4]> var_702 = const()[name = tensor<string, []>("op_702"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_703_cast_fp16 = reshape(shape = var_702, x = linear_30_cast_fp16)[name = tensor<string, []>("op_703_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_234_to_fp16 = const()[name = tensor<string, []>("const_234_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_23_cast_fp16 = mul(x = var_703_cast_fp16, y = const_234_to_fp16)[name = tensor<string, []>("q_23_cast_fp16")];
            tensor<int32, [4]> var_709 = const()[name = tensor<string, []>("op_709"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_710_cast_fp16 = reshape(shape = var_709, x = linear_31_cast_fp16)[name = tensor<string, []>("op_710_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_235_to_fp16 = const()[name = tensor<string, []>("const_235_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_23_cast_fp16 = mul(x = var_710_cast_fp16, y = const_235_to_fp16)[name = tensor<string, []>("k_23_cast_fp16")];
            tensor<int32, [4]> var_716 = const()[name = tensor<string, []>("op_716"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_717_cast_fp16 = reshape(shape = var_716, x = linear_32_cast_fp16)[name = tensor<string, []>("op_717_cast_fp16")];
            tensor<int32, [4]> var_718 = const()[name = tensor<string, []>("op_718"), 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_138_perm_0 = const()[name = tensor<string, []>("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_139_perm_0 = const()[name = tensor<string, []>("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_297 = transpose(perm = transpose_139_perm_0, x = k_23_cast_fp16)[name = tensor<string, []>("transpose_297")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_298 = transpose(perm = transpose_138_perm_0, x = q_23_cast_fp16)[name = tensor<string, []>("transpose_298")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_11_cast_fp16 = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_298, y = transpose_297)[name = tensor<string, []>("qk_11_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_722_cast_fp16 = softmax(axis = var_657, x = qk_11_cast_fp16)[name = tensor<string, []>("op_722_cast_fp16")];
            tensor<bool, []> var_724_transpose_x_0 = const()[name = tensor<string, []>("op_724_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_724_transpose_y_0 = const()[name = tensor<string, []>("op_724_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_299 = transpose(perm = var_718, x = var_717_cast_fp16)[name = tensor<string, []>("transpose_299")];
            tensor<fp16, [1, 20, 1500, 64]> var_724_cast_fp16 = matmul(transpose_x = var_724_transpose_x_0, transpose_y = var_724_transpose_y_0, x = var_722_cast_fp16, y = transpose_299)[name = tensor<string, []>("op_724_cast_fp16")];
            tensor<int32, [4]> var_725 = const()[name = tensor<string, []>("op_725"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_296 = transpose(perm = var_725, x = var_724_cast_fp16)[name = tensor<string, []>("transpose_296")];
            tensor<fp16, [1, 1500, 1280]> x_71_cast_fp16 = reshape(shape = concat_5, x = transpose_296)[name = tensor<string, []>("x_71_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_730_to_fp16 = const()[name = tensor<string, []>("op_730_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221269056)))];
            tensor<fp16, [1280]> var_731_to_fp16 = const()[name = tensor<string, []>("op_731_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224545920)))];
            tensor<fp16, [1, 1500, 1280]> linear_33_cast_fp16 = linear(bias = var_731_to_fp16, weight = var_730_to_fp16, x = x_71_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("x_73_cast_fp16")];
            tensor<int32, [1]> var_738_axes_0 = const()[name = tensor<string, []>("op_738_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224548544)))];
            tensor<fp16, [1280]> blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224551168)))];
            tensor<fp16, [1, 1500, 1280]> var_738_cast_fp16 = layer_norm(axes = var_738_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = tensor<string, []>("op_738_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_747_to_fp16 = const()[name = tensor<string, []>("op_747_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224553792)))];
            tensor<fp16, [5120]> var_748_to_fp16 = const()[name = tensor<string, []>("op_748_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237661056)))];
            tensor<fp16, [1, 1500, 5120]> linear_34_cast_fp16 = linear(bias = var_748_to_fp16, weight = var_747_to_fp16, x = var_738_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
            tensor<string, []> x_77_mode_0 = const()[name = tensor<string, []>("x_77_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = tensor<string, []>("x_77_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_753_to_fp16 = const()[name = tensor<string, []>("op_753_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237671360)))];
            tensor<fp16, [1280]> var_754_to_fp16 = const()[name = tensor<string, []>("op_754_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250778624)))];
            tensor<fp16, [1, 1500, 1280]> linear_35_cast_fp16 = linear(bias = var_754_to_fp16, weight = var_753_to_fp16, x = x_77_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("x_79_cast_fp16")];
            tensor<int32, []> var_764 = const()[name = tensor<string, []>("op_764"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_781_axes_0 = const()[name = tensor<string, []>("op_781_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250781248)))];
            tensor<fp16, [1280]> blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250783872)))];
            tensor<fp16, []> var_770_to_fp16 = const()[name = tensor<string, []>("op_770_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_781_cast_fp16 = layer_norm(axes = var_781_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = tensor<string, []>("op_781_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_792_to_fp16 = const()[name = tensor<string, []>("op_792_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250786496)))];
            tensor<fp16, [1280]> var_793_to_fp16 = const()[name = tensor<string, []>("op_793_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254063360)))];
            tensor<fp16, [1, 1500, 1280]> linear_36_cast_fp16 = linear(bias = var_793_to_fp16, weight = var_792_to_fp16, x = var_781_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_796_to_fp16 = const()[name = tensor<string, []>("op_796_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254065984)))];
            tensor<fp16, [1, 1500, 1280]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_796_to_fp16, x = var_781_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_800_to_fp16 = const()[name = tensor<string, []>("op_800_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257342848)))];
            tensor<fp16, [1280]> var_801_to_fp16 = const()[name = tensor<string, []>("op_801_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(260619712)))];
            tensor<fp16, [1, 1500, 1280]> linear_38_cast_fp16 = linear(bias = var_801_to_fp16, weight = var_800_to_fp16, x = var_781_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<int32, [4]> var_809 = const()[name = tensor<string, []>("op_809"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_810_cast_fp16 = reshape(shape = var_809, x = linear_36_cast_fp16)[name = tensor<string, []>("op_810_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_236_to_fp16 = const()[name = tensor<string, []>("const_236_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_27_cast_fp16 = mul(x = var_810_cast_fp16, y = const_236_to_fp16)[name = tensor<string, []>("q_27_cast_fp16")];
            tensor<int32, [4]> var_816 = const()[name = tensor<string, []>("op_816"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_817_cast_fp16 = reshape(shape = var_816, x = linear_37_cast_fp16)[name = tensor<string, []>("op_817_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_237_to_fp16 = const()[name = tensor<string, []>("const_237_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_27_cast_fp16 = mul(x = var_817_cast_fp16, y = const_237_to_fp16)[name = tensor<string, []>("k_27_cast_fp16")];
            tensor<int32, [4]> var_823 = const()[name = tensor<string, []>("op_823"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_824_cast_fp16 = reshape(shape = var_823, x = linear_38_cast_fp16)[name = tensor<string, []>("op_824_cast_fp16")];
            tensor<int32, [4]> var_825 = const()[name = tensor<string, []>("op_825"), 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_140_perm_0 = const()[name = tensor<string, []>("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_141_perm_0 = const()[name = tensor<string, []>("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_293 = transpose(perm = transpose_141_perm_0, x = k_27_cast_fp16)[name = tensor<string, []>("transpose_293")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_294 = transpose(perm = transpose_140_perm_0, x = q_27_cast_fp16)[name = tensor<string, []>("transpose_294")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_13_cast_fp16 = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_294, y = transpose_293)[name = tensor<string, []>("qk_13_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_829_cast_fp16 = softmax(axis = var_764, x = qk_13_cast_fp16)[name = tensor<string, []>("op_829_cast_fp16")];
            tensor<bool, []> var_831_transpose_x_0 = const()[name = tensor<string, []>("op_831_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_831_transpose_y_0 = const()[name = tensor<string, []>("op_831_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_295 = transpose(perm = var_825, x = var_824_cast_fp16)[name = tensor<string, []>("transpose_295")];
            tensor<fp16, [1, 20, 1500, 64]> var_831_cast_fp16 = matmul(transpose_x = var_831_transpose_x_0, transpose_y = var_831_transpose_y_0, x = var_829_cast_fp16, y = transpose_295)[name = tensor<string, []>("op_831_cast_fp16")];
            tensor<int32, [4]> var_832 = const()[name = tensor<string, []>("op_832"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_292 = transpose(perm = var_832, x = var_831_cast_fp16)[name = tensor<string, []>("transpose_292")];
            tensor<fp16, [1, 1500, 1280]> x_83_cast_fp16 = reshape(shape = concat_6, x = transpose_292)[name = tensor<string, []>("x_83_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_837_to_fp16 = const()[name = tensor<string, []>("op_837_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(260622336)))];
            tensor<fp16, [1280]> var_838_to_fp16 = const()[name = tensor<string, []>("op_838_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263899200)))];
            tensor<fp16, [1, 1500, 1280]> linear_39_cast_fp16 = linear(bias = var_838_to_fp16, weight = var_837_to_fp16, x = x_83_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("x_85_cast_fp16")];
            tensor<int32, [1]> var_845_axes_0 = const()[name = tensor<string, []>("op_845_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263901824)))];
            tensor<fp16, [1280]> blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263904448)))];
            tensor<fp16, [1, 1500, 1280]> var_845_cast_fp16 = layer_norm(axes = var_845_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = tensor<string, []>("op_845_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_854_to_fp16 = const()[name = tensor<string, []>("op_854_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263907072)))];
            tensor<fp16, [5120]> var_855_to_fp16 = const()[name = tensor<string, []>("op_855_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277014336)))];
            tensor<fp16, [1, 1500, 5120]> linear_40_cast_fp16 = linear(bias = var_855_to_fp16, weight = var_854_to_fp16, x = var_845_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
            tensor<string, []> x_89_mode_0 = const()[name = tensor<string, []>("x_89_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = tensor<string, []>("x_89_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_860_to_fp16 = const()[name = tensor<string, []>("op_860_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277024640)))];
            tensor<fp16, [1280]> var_861_to_fp16 = const()[name = tensor<string, []>("op_861_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290131904)))];
            tensor<fp16, [1, 1500, 1280]> linear_41_cast_fp16 = linear(bias = var_861_to_fp16, weight = var_860_to_fp16, x = x_89_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("x_91_cast_fp16")];
            tensor<int32, []> var_871 = const()[name = tensor<string, []>("op_871"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_888_axes_0 = const()[name = tensor<string, []>("op_888_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290134528)))];
            tensor<fp16, [1280]> blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290137152)))];
            tensor<fp16, []> var_877_to_fp16 = const()[name = tensor<string, []>("op_877_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_888_cast_fp16 = layer_norm(axes = var_888_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = tensor<string, []>("op_888_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_899_to_fp16 = const()[name = tensor<string, []>("op_899_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290139776)))];
            tensor<fp16, [1280]> var_900_to_fp16 = const()[name = tensor<string, []>("op_900_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293416640)))];
            tensor<fp16, [1, 1500, 1280]> linear_42_cast_fp16 = linear(bias = var_900_to_fp16, weight = var_899_to_fp16, x = var_888_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_903_to_fp16 = const()[name = tensor<string, []>("op_903_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293419264)))];
            tensor<fp16, [1, 1500, 1280]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_903_to_fp16, x = var_888_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_907_to_fp16 = const()[name = tensor<string, []>("op_907_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296696128)))];
            tensor<fp16, [1280]> var_908_to_fp16 = const()[name = tensor<string, []>("op_908_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299972992)))];
            tensor<fp16, [1, 1500, 1280]> linear_44_cast_fp16 = linear(bias = var_908_to_fp16, weight = var_907_to_fp16, x = var_888_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<int32, [4]> var_916 = const()[name = tensor<string, []>("op_916"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_917_cast_fp16 = reshape(shape = var_916, x = linear_42_cast_fp16)[name = tensor<string, []>("op_917_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_238_to_fp16 = const()[name = tensor<string, []>("const_238_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_31_cast_fp16 = mul(x = var_917_cast_fp16, y = const_238_to_fp16)[name = tensor<string, []>("q_31_cast_fp16")];
            tensor<int32, [4]> var_923 = const()[name = tensor<string, []>("op_923"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_924_cast_fp16 = reshape(shape = var_923, x = linear_43_cast_fp16)[name = tensor<string, []>("op_924_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_239_to_fp16 = const()[name = tensor<string, []>("const_239_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_31_cast_fp16 = mul(x = var_924_cast_fp16, y = const_239_to_fp16)[name = tensor<string, []>("k_31_cast_fp16")];
            tensor<int32, [4]> var_930 = const()[name = tensor<string, []>("op_930"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_931_cast_fp16 = reshape(shape = var_930, x = linear_44_cast_fp16)[name = tensor<string, []>("op_931_cast_fp16")];
            tensor<int32, [4]> var_932 = const()[name = tensor<string, []>("op_932"), 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_142_perm_0 = const()[name = tensor<string, []>("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_143_perm_0 = const()[name = tensor<string, []>("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_289 = transpose(perm = transpose_143_perm_0, x = k_31_cast_fp16)[name = tensor<string, []>("transpose_289")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_290 = transpose(perm = transpose_142_perm_0, x = q_31_cast_fp16)[name = tensor<string, []>("transpose_290")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_15_cast_fp16 = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_290, y = transpose_289)[name = tensor<string, []>("qk_15_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_936_cast_fp16 = softmax(axis = var_871, x = qk_15_cast_fp16)[name = tensor<string, []>("op_936_cast_fp16")];
            tensor<bool, []> var_938_transpose_x_0 = const()[name = tensor<string, []>("op_938_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_938_transpose_y_0 = const()[name = tensor<string, []>("op_938_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_291 = transpose(perm = var_932, x = var_931_cast_fp16)[name = tensor<string, []>("transpose_291")];
            tensor<fp16, [1, 20, 1500, 64]> var_938_cast_fp16 = matmul(transpose_x = var_938_transpose_x_0, transpose_y = var_938_transpose_y_0, x = var_936_cast_fp16, y = transpose_291)[name = tensor<string, []>("op_938_cast_fp16")];
            tensor<int32, [4]> var_939 = const()[name = tensor<string, []>("op_939"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_288 = transpose(perm = var_939, x = var_938_cast_fp16)[name = tensor<string, []>("transpose_288")];
            tensor<fp16, [1, 1500, 1280]> x_95_cast_fp16 = reshape(shape = concat_7, x = transpose_288)[name = tensor<string, []>("x_95_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_944_to_fp16 = const()[name = tensor<string, []>("op_944_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299975616)))];
            tensor<fp16, [1280]> var_945_to_fp16 = const()[name = tensor<string, []>("op_945_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303252480)))];
            tensor<fp16, [1, 1500, 1280]> linear_45_cast_fp16 = linear(bias = var_945_to_fp16, weight = var_944_to_fp16, x = x_95_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("x_97_cast_fp16")];
            tensor<int32, [1]> var_952_axes_0 = const()[name = tensor<string, []>("op_952_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303255104)))];
            tensor<fp16, [1280]> blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303257728)))];
            tensor<fp16, [1, 1500, 1280]> var_952_cast_fp16 = layer_norm(axes = var_952_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = tensor<string, []>("op_952_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_961_to_fp16 = const()[name = tensor<string, []>("op_961_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303260352)))];
            tensor<fp16, [5120]> var_962_to_fp16 = const()[name = tensor<string, []>("op_962_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316367616)))];
            tensor<fp16, [1, 1500, 5120]> linear_46_cast_fp16 = linear(bias = var_962_to_fp16, weight = var_961_to_fp16, x = var_952_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
            tensor<string, []> x_101_mode_0 = const()[name = tensor<string, []>("x_101_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = tensor<string, []>("x_101_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_967_to_fp16 = const()[name = tensor<string, []>("op_967_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316377920)))];
            tensor<fp16, [1280]> var_968_to_fp16 = const()[name = tensor<string, []>("op_968_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329485184)))];
            tensor<fp16, [1, 1500, 1280]> linear_47_cast_fp16 = linear(bias = var_968_to_fp16, weight = var_967_to_fp16, x = x_101_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("x_103_cast_fp16")];
            tensor<int32, []> var_978 = const()[name = tensor<string, []>("op_978"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_995_axes_0 = const()[name = tensor<string, []>("op_995_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329487808)))];
            tensor<fp16, [1280]> blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329490432)))];
            tensor<fp16, []> var_984_to_fp16 = const()[name = tensor<string, []>("op_984_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_995_cast_fp16 = layer_norm(axes = var_995_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = tensor<string, []>("op_995_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1006_to_fp16 = const()[name = tensor<string, []>("op_1006_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329493056)))];
            tensor<fp16, [1280]> var_1007_to_fp16 = const()[name = tensor<string, []>("op_1007_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(332769920)))];
            tensor<fp16, [1, 1500, 1280]> linear_48_cast_fp16 = linear(bias = var_1007_to_fp16, weight = var_1006_to_fp16, x = var_995_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1010_to_fp16 = const()[name = tensor<string, []>("op_1010_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(332772544)))];
            tensor<fp16, [1, 1500, 1280]> linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1010_to_fp16, x = var_995_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1014_to_fp16 = const()[name = tensor<string, []>("op_1014_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(336049408)))];
            tensor<fp16, [1280]> var_1015_to_fp16 = const()[name = tensor<string, []>("op_1015_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339326272)))];
            tensor<fp16, [1, 1500, 1280]> linear_50_cast_fp16 = linear(bias = var_1015_to_fp16, weight = var_1014_to_fp16, x = var_995_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> var_1023 = const()[name = tensor<string, []>("op_1023"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1024_cast_fp16 = reshape(shape = var_1023, x = linear_48_cast_fp16)[name = tensor<string, []>("op_1024_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_240_to_fp16 = const()[name = tensor<string, []>("const_240_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_35_cast_fp16 = mul(x = var_1024_cast_fp16, y = const_240_to_fp16)[name = tensor<string, []>("q_35_cast_fp16")];
            tensor<int32, [4]> var_1030 = const()[name = tensor<string, []>("op_1030"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1031_cast_fp16 = reshape(shape = var_1030, x = linear_49_cast_fp16)[name = tensor<string, []>("op_1031_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_241_to_fp16 = const()[name = tensor<string, []>("const_241_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_35_cast_fp16 = mul(x = var_1031_cast_fp16, y = const_241_to_fp16)[name = tensor<string, []>("k_35_cast_fp16")];
            tensor<int32, [4]> var_1037 = const()[name = tensor<string, []>("op_1037"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1038_cast_fp16 = reshape(shape = var_1037, x = linear_50_cast_fp16)[name = tensor<string, []>("op_1038_cast_fp16")];
            tensor<int32, [4]> var_1039 = const()[name = tensor<string, []>("op_1039"), 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_144_perm_0 = const()[name = tensor<string, []>("transpose_144_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_145_perm_0 = const()[name = tensor<string, []>("transpose_145_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_285 = transpose(perm = transpose_145_perm_0, x = k_35_cast_fp16)[name = tensor<string, []>("transpose_285")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_286 = transpose(perm = transpose_144_perm_0, x = q_35_cast_fp16)[name = tensor<string, []>("transpose_286")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_17_cast_fp16 = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_286, y = transpose_285)[name = tensor<string, []>("qk_17_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1043_cast_fp16 = softmax(axis = var_978, x = qk_17_cast_fp16)[name = tensor<string, []>("op_1043_cast_fp16")];
            tensor<bool, []> var_1045_transpose_x_0 = const()[name = tensor<string, []>("op_1045_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1045_transpose_y_0 = const()[name = tensor<string, []>("op_1045_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_287 = transpose(perm = var_1039, x = var_1038_cast_fp16)[name = tensor<string, []>("transpose_287")];
            tensor<fp16, [1, 20, 1500, 64]> var_1045_cast_fp16 = matmul(transpose_x = var_1045_transpose_x_0, transpose_y = var_1045_transpose_y_0, x = var_1043_cast_fp16, y = transpose_287)[name = tensor<string, []>("op_1045_cast_fp16")];
            tensor<int32, [4]> var_1046 = const()[name = tensor<string, []>("op_1046"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_284 = transpose(perm = var_1046, x = var_1045_cast_fp16)[name = tensor<string, []>("transpose_284")];
            tensor<fp16, [1, 1500, 1280]> x_107_cast_fp16 = reshape(shape = concat_8, x = transpose_284)[name = tensor<string, []>("x_107_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1051_to_fp16 = const()[name = tensor<string, []>("op_1051_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339328896)))];
            tensor<fp16, [1280]> var_1052_to_fp16 = const()[name = tensor<string, []>("op_1052_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342605760)))];
            tensor<fp16, [1, 1500, 1280]> linear_51_cast_fp16 = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_107_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("x_109_cast_fp16")];
            tensor<int32, [1]> var_1059_axes_0 = const()[name = tensor<string, []>("op_1059_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342608384)))];
            tensor<fp16, [1280]> blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342611008)))];
            tensor<fp16, [1, 1500, 1280]> var_1059_cast_fp16 = layer_norm(axes = var_1059_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = tensor<string, []>("op_1059_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1068_to_fp16 = const()[name = tensor<string, []>("op_1068_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342613632)))];
            tensor<fp16, [5120]> var_1069_to_fp16 = const()[name = tensor<string, []>("op_1069_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355720896)))];
            tensor<fp16, [1, 1500, 5120]> linear_52_cast_fp16 = linear(bias = var_1069_to_fp16, weight = var_1068_to_fp16, x = var_1059_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
            tensor<string, []> x_113_mode_0 = const()[name = tensor<string, []>("x_113_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = tensor<string, []>("x_113_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1074_to_fp16 = const()[name = tensor<string, []>("op_1074_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355731200)))];
            tensor<fp16, [1280]> var_1075_to_fp16 = const()[name = tensor<string, []>("op_1075_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368838464)))];
            tensor<fp16, [1, 1500, 1280]> linear_53_cast_fp16 = linear(bias = var_1075_to_fp16, weight = var_1074_to_fp16, x = x_113_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("x_115_cast_fp16")];
            tensor<int32, []> var_1085 = const()[name = tensor<string, []>("op_1085"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1102_axes_0 = const()[name = tensor<string, []>("op_1102_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368841088)))];
            tensor<fp16, [1280]> blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368843712)))];
            tensor<fp16, []> var_1091_to_fp16 = const()[name = tensor<string, []>("op_1091_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1102_cast_fp16 = layer_norm(axes = var_1102_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = tensor<string, []>("op_1102_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1113_to_fp16 = const()[name = tensor<string, []>("op_1113_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368846336)))];
            tensor<fp16, [1280]> var_1114_to_fp16 = const()[name = tensor<string, []>("op_1114_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(372123200)))];
            tensor<fp16, [1, 1500, 1280]> linear_54_cast_fp16 = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1102_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1117_to_fp16 = const()[name = tensor<string, []>("op_1117_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(372125824)))];
            tensor<fp16, [1, 1500, 1280]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1117_to_fp16, x = var_1102_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1121_to_fp16 = const()[name = tensor<string, []>("op_1121_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375402688)))];
            tensor<fp16, [1280]> var_1122_to_fp16 = const()[name = tensor<string, []>("op_1122_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378679552)))];
            tensor<fp16, [1, 1500, 1280]> linear_56_cast_fp16 = linear(bias = var_1122_to_fp16, weight = var_1121_to_fp16, x = var_1102_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<int32, [4]> var_1130 = const()[name = tensor<string, []>("op_1130"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1131_cast_fp16 = reshape(shape = var_1130, x = linear_54_cast_fp16)[name = tensor<string, []>("op_1131_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_242_to_fp16 = const()[name = tensor<string, []>("const_242_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_39_cast_fp16 = mul(x = var_1131_cast_fp16, y = const_242_to_fp16)[name = tensor<string, []>("q_39_cast_fp16")];
            tensor<int32, [4]> var_1137 = const()[name = tensor<string, []>("op_1137"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1138_cast_fp16 = reshape(shape = var_1137, x = linear_55_cast_fp16)[name = tensor<string, []>("op_1138_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_243_to_fp16 = const()[name = tensor<string, []>("const_243_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_39_cast_fp16 = mul(x = var_1138_cast_fp16, y = const_243_to_fp16)[name = tensor<string, []>("k_39_cast_fp16")];
            tensor<int32, [4]> var_1144 = const()[name = tensor<string, []>("op_1144"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1145_cast_fp16 = reshape(shape = var_1144, x = linear_56_cast_fp16)[name = tensor<string, []>("op_1145_cast_fp16")];
            tensor<int32, [4]> var_1146 = const()[name = tensor<string, []>("op_1146"), 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_146_perm_0 = const()[name = tensor<string, []>("transpose_146_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_147_perm_0 = const()[name = tensor<string, []>("transpose_147_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_281 = transpose(perm = transpose_147_perm_0, x = k_39_cast_fp16)[name = tensor<string, []>("transpose_281")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_282 = transpose(perm = transpose_146_perm_0, x = q_39_cast_fp16)[name = tensor<string, []>("transpose_282")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_19_cast_fp16 = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_282, y = transpose_281)[name = tensor<string, []>("qk_19_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1150_cast_fp16 = softmax(axis = var_1085, x = qk_19_cast_fp16)[name = tensor<string, []>("op_1150_cast_fp16")];
            tensor<bool, []> var_1152_transpose_x_0 = const()[name = tensor<string, []>("op_1152_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1152_transpose_y_0 = const()[name = tensor<string, []>("op_1152_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_283 = transpose(perm = var_1146, x = var_1145_cast_fp16)[name = tensor<string, []>("transpose_283")];
            tensor<fp16, [1, 20, 1500, 64]> var_1152_cast_fp16 = matmul(transpose_x = var_1152_transpose_x_0, transpose_y = var_1152_transpose_y_0, x = var_1150_cast_fp16, y = transpose_283)[name = tensor<string, []>("op_1152_cast_fp16")];
            tensor<int32, [4]> var_1153 = const()[name = tensor<string, []>("op_1153"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_280 = transpose(perm = var_1153, x = var_1152_cast_fp16)[name = tensor<string, []>("transpose_280")];
            tensor<fp16, [1, 1500, 1280]> x_119_cast_fp16 = reshape(shape = concat_9, x = transpose_280)[name = tensor<string, []>("x_119_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1158_to_fp16 = const()[name = tensor<string, []>("op_1158_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378682176)))];
            tensor<fp16, [1280]> var_1159_to_fp16 = const()[name = tensor<string, []>("op_1159_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381959040)))];
            tensor<fp16, [1, 1500, 1280]> linear_57_cast_fp16 = linear(bias = var_1159_to_fp16, weight = var_1158_to_fp16, x = x_119_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("x_121_cast_fp16")];
            tensor<int32, [1]> var_1166_axes_0 = const()[name = tensor<string, []>("op_1166_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381961664)))];
            tensor<fp16, [1280]> blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381964288)))];
            tensor<fp16, [1, 1500, 1280]> var_1166_cast_fp16 = layer_norm(axes = var_1166_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = tensor<string, []>("op_1166_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1175_to_fp16 = const()[name = tensor<string, []>("op_1175_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381966912)))];
            tensor<fp16, [5120]> var_1176_to_fp16 = const()[name = tensor<string, []>("op_1176_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(395074176)))];
            tensor<fp16, [1, 1500, 5120]> linear_58_cast_fp16 = linear(bias = var_1176_to_fp16, weight = var_1175_to_fp16, x = var_1166_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
            tensor<string, []> x_125_mode_0 = const()[name = tensor<string, []>("x_125_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = tensor<string, []>("x_125_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1181_to_fp16 = const()[name = tensor<string, []>("op_1181_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(395084480)))];
            tensor<fp16, [1280]> var_1182_to_fp16 = const()[name = tensor<string, []>("op_1182_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408191744)))];
            tensor<fp16, [1, 1500, 1280]> linear_59_cast_fp16 = linear(bias = var_1182_to_fp16, weight = var_1181_to_fp16, x = x_125_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("x_127_cast_fp16")];
            tensor<int32, []> var_1192 = const()[name = tensor<string, []>("op_1192"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1209_axes_0 = const()[name = tensor<string, []>("op_1209_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408194368)))];
            tensor<fp16, [1280]> blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408196992)))];
            tensor<fp16, []> var_1198_to_fp16 = const()[name = tensor<string, []>("op_1198_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1209_cast_fp16 = layer_norm(axes = var_1209_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = tensor<string, []>("op_1209_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1220_to_fp16 = const()[name = tensor<string, []>("op_1220_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408199616)))];
            tensor<fp16, [1280]> var_1221_to_fp16 = const()[name = tensor<string, []>("op_1221_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411476480)))];
            tensor<fp16, [1, 1500, 1280]> linear_60_cast_fp16 = linear(bias = var_1221_to_fp16, weight = var_1220_to_fp16, x = var_1209_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1224_to_fp16 = const()[name = tensor<string, []>("op_1224_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411479104)))];
            tensor<fp16, [1, 1500, 1280]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1224_to_fp16, x = var_1209_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1228_to_fp16 = const()[name = tensor<string, []>("op_1228_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(414755968)))];
            tensor<fp16, [1280]> var_1229_to_fp16 = const()[name = tensor<string, []>("op_1229_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(418032832)))];
            tensor<fp16, [1, 1500, 1280]> linear_62_cast_fp16 = linear(bias = var_1229_to_fp16, weight = var_1228_to_fp16, x = var_1209_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<int32, [4]> var_1237 = const()[name = tensor<string, []>("op_1237"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1238_cast_fp16 = reshape(shape = var_1237, x = linear_60_cast_fp16)[name = tensor<string, []>("op_1238_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_244_to_fp16 = const()[name = tensor<string, []>("const_244_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_43_cast_fp16 = mul(x = var_1238_cast_fp16, y = const_244_to_fp16)[name = tensor<string, []>("q_43_cast_fp16")];
            tensor<int32, [4]> var_1244 = const()[name = tensor<string, []>("op_1244"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1245_cast_fp16 = reshape(shape = var_1244, x = linear_61_cast_fp16)[name = tensor<string, []>("op_1245_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_245_to_fp16 = const()[name = tensor<string, []>("const_245_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_43_cast_fp16 = mul(x = var_1245_cast_fp16, y = const_245_to_fp16)[name = tensor<string, []>("k_43_cast_fp16")];
            tensor<int32, [4]> var_1251 = const()[name = tensor<string, []>("op_1251"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1252_cast_fp16 = reshape(shape = var_1251, x = linear_62_cast_fp16)[name = tensor<string, []>("op_1252_cast_fp16")];
            tensor<int32, [4]> var_1253 = const()[name = tensor<string, []>("op_1253"), 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_148_perm_0 = const()[name = tensor<string, []>("transpose_148_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_149_perm_0 = const()[name = tensor<string, []>("transpose_149_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_277 = transpose(perm = transpose_149_perm_0, x = k_43_cast_fp16)[name = tensor<string, []>("transpose_277")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_278 = transpose(perm = transpose_148_perm_0, x = q_43_cast_fp16)[name = tensor<string, []>("transpose_278")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_21_cast_fp16 = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_278, y = transpose_277)[name = tensor<string, []>("qk_21_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1257_cast_fp16 = softmax(axis = var_1192, x = qk_21_cast_fp16)[name = tensor<string, []>("op_1257_cast_fp16")];
            tensor<bool, []> var_1259_transpose_x_0 = const()[name = tensor<string, []>("op_1259_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1259_transpose_y_0 = const()[name = tensor<string, []>("op_1259_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_279 = transpose(perm = var_1253, x = var_1252_cast_fp16)[name = tensor<string, []>("transpose_279")];
            tensor<fp16, [1, 20, 1500, 64]> var_1259_cast_fp16 = matmul(transpose_x = var_1259_transpose_x_0, transpose_y = var_1259_transpose_y_0, x = var_1257_cast_fp16, y = transpose_279)[name = tensor<string, []>("op_1259_cast_fp16")];
            tensor<int32, [4]> var_1260 = const()[name = tensor<string, []>("op_1260"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_276 = transpose(perm = var_1260, x = var_1259_cast_fp16)[name = tensor<string, []>("transpose_276")];
            tensor<fp16, [1, 1500, 1280]> x_131_cast_fp16 = reshape(shape = concat_10, x = transpose_276)[name = tensor<string, []>("x_131_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1265_to_fp16 = const()[name = tensor<string, []>("op_1265_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(418035456)))];
            tensor<fp16, [1280]> var_1266_to_fp16 = const()[name = tensor<string, []>("op_1266_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421312320)))];
            tensor<fp16, [1, 1500, 1280]> linear_63_cast_fp16 = linear(bias = var_1266_to_fp16, weight = var_1265_to_fp16, x = x_131_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("x_133_cast_fp16")];
            tensor<int32, [1]> var_1273_axes_0 = const()[name = tensor<string, []>("op_1273_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421314944)))];
            tensor<fp16, [1280]> blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421317568)))];
            tensor<fp16, [1, 1500, 1280]> var_1273_cast_fp16 = layer_norm(axes = var_1273_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = tensor<string, []>("op_1273_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1282_to_fp16 = const()[name = tensor<string, []>("op_1282_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(421320192)))];
            tensor<fp16, [5120]> var_1283_to_fp16 = const()[name = tensor<string, []>("op_1283_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434427456)))];
            tensor<fp16, [1, 1500, 5120]> linear_64_cast_fp16 = linear(bias = var_1283_to_fp16, weight = var_1282_to_fp16, x = var_1273_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
            tensor<string, []> x_137_mode_0 = const()[name = tensor<string, []>("x_137_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = tensor<string, []>("x_137_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1288_to_fp16 = const()[name = tensor<string, []>("op_1288_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434437760)))];
            tensor<fp16, [1280]> var_1289_to_fp16 = const()[name = tensor<string, []>("op_1289_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447545024)))];
            tensor<fp16, [1, 1500, 1280]> linear_65_cast_fp16 = linear(bias = var_1289_to_fp16, weight = var_1288_to_fp16, x = x_137_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("x_139_cast_fp16")];
            tensor<int32, []> var_1299 = const()[name = tensor<string, []>("op_1299"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1316_axes_0 = const()[name = tensor<string, []>("op_1316_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447547648)))];
            tensor<fp16, [1280]> blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447550272)))];
            tensor<fp16, []> var_1305_to_fp16 = const()[name = tensor<string, []>("op_1305_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1316_cast_fp16 = layer_norm(axes = var_1316_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = tensor<string, []>("op_1316_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1327_to_fp16 = const()[name = tensor<string, []>("op_1327_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447552896)))];
            tensor<fp16, [1280]> var_1328_to_fp16 = const()[name = tensor<string, []>("op_1328_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450829760)))];
            tensor<fp16, [1, 1500, 1280]> linear_66_cast_fp16 = linear(bias = var_1328_to_fp16, weight = var_1327_to_fp16, x = var_1316_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1331_to_fp16 = const()[name = tensor<string, []>("op_1331_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450832384)))];
            tensor<fp16, [1, 1500, 1280]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1331_to_fp16, x = var_1316_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1335_to_fp16 = const()[name = tensor<string, []>("op_1335_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454109248)))];
            tensor<fp16, [1280]> var_1336_to_fp16 = const()[name = tensor<string, []>("op_1336_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457386112)))];
            tensor<fp16, [1, 1500, 1280]> linear_68_cast_fp16 = linear(bias = var_1336_to_fp16, weight = var_1335_to_fp16, x = var_1316_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<int32, [4]> var_1344 = const()[name = tensor<string, []>("op_1344"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1345_cast_fp16 = reshape(shape = var_1344, x = linear_66_cast_fp16)[name = tensor<string, []>("op_1345_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_246_to_fp16 = const()[name = tensor<string, []>("const_246_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_47_cast_fp16 = mul(x = var_1345_cast_fp16, y = const_246_to_fp16)[name = tensor<string, []>("q_47_cast_fp16")];
            tensor<int32, [4]> var_1351 = const()[name = tensor<string, []>("op_1351"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1352_cast_fp16 = reshape(shape = var_1351, x = linear_67_cast_fp16)[name = tensor<string, []>("op_1352_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_247_to_fp16 = const()[name = tensor<string, []>("const_247_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_47_cast_fp16 = mul(x = var_1352_cast_fp16, y = const_247_to_fp16)[name = tensor<string, []>("k_47_cast_fp16")];
            tensor<int32, [4]> var_1358 = const()[name = tensor<string, []>("op_1358"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1359_cast_fp16 = reshape(shape = var_1358, x = linear_68_cast_fp16)[name = tensor<string, []>("op_1359_cast_fp16")];
            tensor<int32, [4]> var_1360 = const()[name = tensor<string, []>("op_1360"), 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_150_perm_0 = const()[name = tensor<string, []>("transpose_150_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_151_perm_0 = const()[name = tensor<string, []>("transpose_151_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_273 = transpose(perm = transpose_151_perm_0, x = k_47_cast_fp16)[name = tensor<string, []>("transpose_273")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_274 = transpose(perm = transpose_150_perm_0, x = q_47_cast_fp16)[name = tensor<string, []>("transpose_274")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_23_cast_fp16 = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_274, y = transpose_273)[name = tensor<string, []>("qk_23_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1364_cast_fp16 = softmax(axis = var_1299, x = qk_23_cast_fp16)[name = tensor<string, []>("op_1364_cast_fp16")];
            tensor<bool, []> var_1366_transpose_x_0 = const()[name = tensor<string, []>("op_1366_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1366_transpose_y_0 = const()[name = tensor<string, []>("op_1366_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_275 = transpose(perm = var_1360, x = var_1359_cast_fp16)[name = tensor<string, []>("transpose_275")];
            tensor<fp16, [1, 20, 1500, 64]> var_1366_cast_fp16 = matmul(transpose_x = var_1366_transpose_x_0, transpose_y = var_1366_transpose_y_0, x = var_1364_cast_fp16, y = transpose_275)[name = tensor<string, []>("op_1366_cast_fp16")];
            tensor<int32, [4]> var_1367 = const()[name = tensor<string, []>("op_1367"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_272 = transpose(perm = var_1367, x = var_1366_cast_fp16)[name = tensor<string, []>("transpose_272")];
            tensor<fp16, [1, 1500, 1280]> x_143_cast_fp16 = reshape(shape = concat_11, x = transpose_272)[name = tensor<string, []>("x_143_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1372_to_fp16 = const()[name = tensor<string, []>("op_1372_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457388736)))];
            tensor<fp16, [1280]> var_1373_to_fp16 = const()[name = tensor<string, []>("op_1373_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460665600)))];
            tensor<fp16, [1, 1500, 1280]> linear_69_cast_fp16 = linear(bias = var_1373_to_fp16, weight = var_1372_to_fp16, x = x_143_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("x_145_cast_fp16")];
            tensor<int32, [1]> var_1380_axes_0 = const()[name = tensor<string, []>("op_1380_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460668224)))];
            tensor<fp16, [1280]> blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460670848)))];
            tensor<fp16, [1, 1500, 1280]> var_1380_cast_fp16 = layer_norm(axes = var_1380_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = tensor<string, []>("op_1380_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1389_to_fp16 = const()[name = tensor<string, []>("op_1389_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460673472)))];
            tensor<fp16, [5120]> var_1390_to_fp16 = const()[name = tensor<string, []>("op_1390_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473780736)))];
            tensor<fp16, [1, 1500, 5120]> linear_70_cast_fp16 = linear(bias = var_1390_to_fp16, weight = var_1389_to_fp16, x = var_1380_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
            tensor<string, []> x_149_mode_0 = const()[name = tensor<string, []>("x_149_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = tensor<string, []>("x_149_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1395_to_fp16 = const()[name = tensor<string, []>("op_1395_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473791040)))];
            tensor<fp16, [1280]> var_1396_to_fp16 = const()[name = tensor<string, []>("op_1396_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486898304)))];
            tensor<fp16, [1, 1500, 1280]> linear_71_cast_fp16 = linear(bias = var_1396_to_fp16, weight = var_1395_to_fp16, x = x_149_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_151_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("x_151_cast_fp16")];
            tensor<int32, []> var_1406 = const()[name = tensor<string, []>("op_1406"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1423_axes_0 = const()[name = tensor<string, []>("op_1423_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_12_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486900928)))];
            tensor<fp16, [1280]> blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486903552)))];
            tensor<fp16, []> var_1412_to_fp16 = const()[name = tensor<string, []>("op_1412_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1423_cast_fp16 = layer_norm(axes = var_1423_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = tensor<string, []>("op_1423_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1434_to_fp16 = const()[name = tensor<string, []>("op_1434_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486906176)))];
            tensor<fp16, [1280]> var_1435_to_fp16 = const()[name = tensor<string, []>("op_1435_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490183040)))];
            tensor<fp16, [1, 1500, 1280]> linear_72_cast_fp16 = linear(bias = var_1435_to_fp16, weight = var_1434_to_fp16, x = var_1423_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1438_to_fp16 = const()[name = tensor<string, []>("op_1438_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490185664)))];
            tensor<fp16, [1, 1500, 1280]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1438_to_fp16, x = var_1423_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1442_to_fp16 = const()[name = tensor<string, []>("op_1442_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493462528)))];
            tensor<fp16, [1280]> var_1443_to_fp16 = const()[name = tensor<string, []>("op_1443_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496739392)))];
            tensor<fp16, [1, 1500, 1280]> linear_74_cast_fp16 = linear(bias = var_1443_to_fp16, weight = var_1442_to_fp16, x = var_1423_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
            tensor<int32, [4]> var_1451 = const()[name = tensor<string, []>("op_1451"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1452_cast_fp16 = reshape(shape = var_1451, x = linear_72_cast_fp16)[name = tensor<string, []>("op_1452_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_248_to_fp16 = const()[name = tensor<string, []>("const_248_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_51_cast_fp16 = mul(x = var_1452_cast_fp16, y = const_248_to_fp16)[name = tensor<string, []>("q_51_cast_fp16")];
            tensor<int32, [4]> var_1458 = const()[name = tensor<string, []>("op_1458"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1459_cast_fp16 = reshape(shape = var_1458, x = linear_73_cast_fp16)[name = tensor<string, []>("op_1459_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_249_to_fp16 = const()[name = tensor<string, []>("const_249_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_51_cast_fp16 = mul(x = var_1459_cast_fp16, y = const_249_to_fp16)[name = tensor<string, []>("k_51_cast_fp16")];
            tensor<int32, [4]> var_1465 = const()[name = tensor<string, []>("op_1465"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1466_cast_fp16 = reshape(shape = var_1465, x = linear_74_cast_fp16)[name = tensor<string, []>("op_1466_cast_fp16")];
            tensor<int32, [4]> var_1467 = const()[name = tensor<string, []>("op_1467"), 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_152_perm_0 = const()[name = tensor<string, []>("transpose_152_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_153_perm_0 = const()[name = tensor<string, []>("transpose_153_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_269 = transpose(perm = transpose_153_perm_0, x = k_51_cast_fp16)[name = tensor<string, []>("transpose_269")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_270 = transpose(perm = transpose_152_perm_0, x = q_51_cast_fp16)[name = tensor<string, []>("transpose_270")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_25_cast_fp16 = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_270, y = transpose_269)[name = tensor<string, []>("qk_25_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1471_cast_fp16 = softmax(axis = var_1406, x = qk_25_cast_fp16)[name = tensor<string, []>("op_1471_cast_fp16")];
            tensor<bool, []> var_1473_transpose_x_0 = const()[name = tensor<string, []>("op_1473_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1473_transpose_y_0 = const()[name = tensor<string, []>("op_1473_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_271 = transpose(perm = var_1467, x = var_1466_cast_fp16)[name = tensor<string, []>("transpose_271")];
            tensor<fp16, [1, 20, 1500, 64]> var_1473_cast_fp16 = matmul(transpose_x = var_1473_transpose_x_0, transpose_y = var_1473_transpose_y_0, x = var_1471_cast_fp16, y = transpose_271)[name = tensor<string, []>("op_1473_cast_fp16")];
            tensor<int32, [4]> var_1474 = const()[name = tensor<string, []>("op_1474"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_268 = transpose(perm = var_1474, x = var_1473_cast_fp16)[name = tensor<string, []>("transpose_268")];
            tensor<fp16, [1, 1500, 1280]> x_155_cast_fp16 = reshape(shape = concat_12, x = transpose_268)[name = tensor<string, []>("x_155_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1479_to_fp16 = const()[name = tensor<string, []>("op_1479_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496742016)))];
            tensor<fp16, [1280]> var_1480_to_fp16 = const()[name = tensor<string, []>("op_1480_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500018880)))];
            tensor<fp16, [1, 1500, 1280]> linear_75_cast_fp16 = linear(bias = var_1480_to_fp16, weight = var_1479_to_fp16, x = x_155_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_157_cast_fp16 = add(x = x_151_cast_fp16, y = linear_75_cast_fp16)[name = tensor<string, []>("x_157_cast_fp16")];
            tensor<int32, [1]> var_1487_axes_0 = const()[name = tensor<string, []>("op_1487_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_12_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500021504)))];
            tensor<fp16, [1280]> blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500024128)))];
            tensor<fp16, [1, 1500, 1280]> var_1487_cast_fp16 = layer_norm(axes = var_1487_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = tensor<string, []>("op_1487_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1496_to_fp16 = const()[name = tensor<string, []>("op_1496_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500026752)))];
            tensor<fp16, [5120]> var_1497_to_fp16 = const()[name = tensor<string, []>("op_1497_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513134016)))];
            tensor<fp16, [1, 1500, 5120]> linear_76_cast_fp16 = linear(bias = var_1497_to_fp16, weight = var_1496_to_fp16, x = var_1487_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")];
            tensor<string, []> x_161_mode_0 = const()[name = tensor<string, []>("x_161_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_161_cast_fp16 = gelu(mode = x_161_mode_0, x = linear_76_cast_fp16)[name = tensor<string, []>("x_161_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1502_to_fp16 = const()[name = tensor<string, []>("op_1502_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513144320)))];
            tensor<fp16, [1280]> var_1503_to_fp16 = const()[name = tensor<string, []>("op_1503_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526251584)))];
            tensor<fp16, [1, 1500, 1280]> linear_77_cast_fp16 = linear(bias = var_1503_to_fp16, weight = var_1502_to_fp16, x = x_161_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_163_cast_fp16 = add(x = x_157_cast_fp16, y = linear_77_cast_fp16)[name = tensor<string, []>("x_163_cast_fp16")];
            tensor<int32, []> var_1513 = const()[name = tensor<string, []>("op_1513"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1530_axes_0 = const()[name = tensor<string, []>("op_1530_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_13_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526254208)))];
            tensor<fp16, [1280]> blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526256832)))];
            tensor<fp16, []> var_1519_to_fp16 = const()[name = tensor<string, []>("op_1519_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1530_cast_fp16 = layer_norm(axes = var_1530_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = tensor<string, []>("op_1530_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1541_to_fp16 = const()[name = tensor<string, []>("op_1541_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526259456)))];
            tensor<fp16, [1280]> var_1542_to_fp16 = const()[name = tensor<string, []>("op_1542_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(529536320)))];
            tensor<fp16, [1, 1500, 1280]> linear_78_cast_fp16 = linear(bias = var_1542_to_fp16, weight = var_1541_to_fp16, x = var_1530_cast_fp16)[name = tensor<string, []>("linear_78_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1545_to_fp16 = const()[name = tensor<string, []>("op_1545_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(529538944)))];
            tensor<fp16, [1, 1500, 1280]> linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1545_to_fp16, x = var_1530_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1549_to_fp16 = const()[name = tensor<string, []>("op_1549_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(532815808)))];
            tensor<fp16, [1280]> var_1550_to_fp16 = const()[name = tensor<string, []>("op_1550_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(536092672)))];
            tensor<fp16, [1, 1500, 1280]> linear_80_cast_fp16 = linear(bias = var_1550_to_fp16, weight = var_1549_to_fp16, x = var_1530_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")];
            tensor<int32, [4]> var_1558 = const()[name = tensor<string, []>("op_1558"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1559_cast_fp16 = reshape(shape = var_1558, x = linear_78_cast_fp16)[name = tensor<string, []>("op_1559_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_250_to_fp16 = const()[name = tensor<string, []>("const_250_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_55_cast_fp16 = mul(x = var_1559_cast_fp16, y = const_250_to_fp16)[name = tensor<string, []>("q_55_cast_fp16")];
            tensor<int32, [4]> var_1565 = const()[name = tensor<string, []>("op_1565"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1566_cast_fp16 = reshape(shape = var_1565, x = linear_79_cast_fp16)[name = tensor<string, []>("op_1566_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_251_to_fp16 = const()[name = tensor<string, []>("const_251_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_55_cast_fp16 = mul(x = var_1566_cast_fp16, y = const_251_to_fp16)[name = tensor<string, []>("k_55_cast_fp16")];
            tensor<int32, [4]> var_1572 = const()[name = tensor<string, []>("op_1572"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1573_cast_fp16 = reshape(shape = var_1572, x = linear_80_cast_fp16)[name = tensor<string, []>("op_1573_cast_fp16")];
            tensor<int32, [4]> var_1574 = const()[name = tensor<string, []>("op_1574"), 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_154_perm_0 = const()[name = tensor<string, []>("transpose_154_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_155_perm_0 = const()[name = tensor<string, []>("transpose_155_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_265 = transpose(perm = transpose_155_perm_0, x = k_55_cast_fp16)[name = tensor<string, []>("transpose_265")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_266 = transpose(perm = transpose_154_perm_0, x = q_55_cast_fp16)[name = tensor<string, []>("transpose_266")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_27_cast_fp16 = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_266, y = transpose_265)[name = tensor<string, []>("qk_27_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1578_cast_fp16 = softmax(axis = var_1513, x = qk_27_cast_fp16)[name = tensor<string, []>("op_1578_cast_fp16")];
            tensor<bool, []> var_1580_transpose_x_0 = const()[name = tensor<string, []>("op_1580_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1580_transpose_y_0 = const()[name = tensor<string, []>("op_1580_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_267 = transpose(perm = var_1574, x = var_1573_cast_fp16)[name = tensor<string, []>("transpose_267")];
            tensor<fp16, [1, 20, 1500, 64]> var_1580_cast_fp16 = matmul(transpose_x = var_1580_transpose_x_0, transpose_y = var_1580_transpose_y_0, x = var_1578_cast_fp16, y = transpose_267)[name = tensor<string, []>("op_1580_cast_fp16")];
            tensor<int32, [4]> var_1581 = const()[name = tensor<string, []>("op_1581"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_264 = transpose(perm = var_1581, x = var_1580_cast_fp16)[name = tensor<string, []>("transpose_264")];
            tensor<fp16, [1, 1500, 1280]> x_167_cast_fp16 = reshape(shape = concat_13, x = transpose_264)[name = tensor<string, []>("x_167_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1586_to_fp16 = const()[name = tensor<string, []>("op_1586_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(536095296)))];
            tensor<fp16, [1280]> var_1587_to_fp16 = const()[name = tensor<string, []>("op_1587_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539372160)))];
            tensor<fp16, [1, 1500, 1280]> linear_81_cast_fp16 = linear(bias = var_1587_to_fp16, weight = var_1586_to_fp16, x = x_167_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_169_cast_fp16 = add(x = x_163_cast_fp16, y = linear_81_cast_fp16)[name = tensor<string, []>("x_169_cast_fp16")];
            tensor<int32, [1]> var_1594_axes_0 = const()[name = tensor<string, []>("op_1594_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_13_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539374784)))];
            tensor<fp16, [1280]> blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539377408)))];
            tensor<fp16, [1, 1500, 1280]> var_1594_cast_fp16 = layer_norm(axes = var_1594_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = tensor<string, []>("op_1594_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1603_to_fp16 = const()[name = tensor<string, []>("op_1603_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539380032)))];
            tensor<fp16, [5120]> var_1604_to_fp16 = const()[name = tensor<string, []>("op_1604_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552487296)))];
            tensor<fp16, [1, 1500, 5120]> linear_82_cast_fp16 = linear(bias = var_1604_to_fp16, weight = var_1603_to_fp16, x = var_1594_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")];
            tensor<string, []> x_173_mode_0 = const()[name = tensor<string, []>("x_173_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_173_cast_fp16 = gelu(mode = x_173_mode_0, x = linear_82_cast_fp16)[name = tensor<string, []>("x_173_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1609_to_fp16 = const()[name = tensor<string, []>("op_1609_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552497600)))];
            tensor<fp16, [1280]> var_1610_to_fp16 = const()[name = tensor<string, []>("op_1610_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565604864)))];
            tensor<fp16, [1, 1500, 1280]> linear_83_cast_fp16 = linear(bias = var_1610_to_fp16, weight = var_1609_to_fp16, x = x_173_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_175_cast_fp16 = add(x = x_169_cast_fp16, y = linear_83_cast_fp16)[name = tensor<string, []>("x_175_cast_fp16")];
            tensor<int32, []> var_1620 = const()[name = tensor<string, []>("op_1620"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1637_axes_0 = const()[name = tensor<string, []>("op_1637_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_14_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565607488)))];
            tensor<fp16, [1280]> blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565610112)))];
            tensor<fp16, []> var_1626_to_fp16 = const()[name = tensor<string, []>("op_1626_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1637_cast_fp16 = layer_norm(axes = var_1637_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = tensor<string, []>("op_1637_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1648_to_fp16 = const()[name = tensor<string, []>("op_1648_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565612736)))];
            tensor<fp16, [1280]> var_1649_to_fp16 = const()[name = tensor<string, []>("op_1649_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568889600)))];
            tensor<fp16, [1, 1500, 1280]> linear_84_cast_fp16 = linear(bias = var_1649_to_fp16, weight = var_1648_to_fp16, x = var_1637_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1652_to_fp16 = const()[name = tensor<string, []>("op_1652_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568892224)))];
            tensor<fp16, [1, 1500, 1280]> linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1652_to_fp16, x = var_1637_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1656_to_fp16 = const()[name = tensor<string, []>("op_1656_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572169088)))];
            tensor<fp16, [1280]> var_1657_to_fp16 = const()[name = tensor<string, []>("op_1657_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575445952)))];
            tensor<fp16, [1, 1500, 1280]> linear_86_cast_fp16 = linear(bias = var_1657_to_fp16, weight = var_1656_to_fp16, x = var_1637_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")];
            tensor<int32, [4]> var_1665 = const()[name = tensor<string, []>("op_1665"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1666_cast_fp16 = reshape(shape = var_1665, x = linear_84_cast_fp16)[name = tensor<string, []>("op_1666_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_252_to_fp16 = const()[name = tensor<string, []>("const_252_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_59_cast_fp16 = mul(x = var_1666_cast_fp16, y = const_252_to_fp16)[name = tensor<string, []>("q_59_cast_fp16")];
            tensor<int32, [4]> var_1672 = const()[name = tensor<string, []>("op_1672"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1673_cast_fp16 = reshape(shape = var_1672, x = linear_85_cast_fp16)[name = tensor<string, []>("op_1673_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_253_to_fp16 = const()[name = tensor<string, []>("const_253_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_59_cast_fp16 = mul(x = var_1673_cast_fp16, y = const_253_to_fp16)[name = tensor<string, []>("k_59_cast_fp16")];
            tensor<int32, [4]> var_1679 = const()[name = tensor<string, []>("op_1679"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1680_cast_fp16 = reshape(shape = var_1679, x = linear_86_cast_fp16)[name = tensor<string, []>("op_1680_cast_fp16")];
            tensor<int32, [4]> var_1681 = const()[name = tensor<string, []>("op_1681"), 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_156_perm_0 = const()[name = tensor<string, []>("transpose_156_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_157_perm_0 = const()[name = tensor<string, []>("transpose_157_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_261 = transpose(perm = transpose_157_perm_0, x = k_59_cast_fp16)[name = tensor<string, []>("transpose_261")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_262 = transpose(perm = transpose_156_perm_0, x = q_59_cast_fp16)[name = tensor<string, []>("transpose_262")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_29_cast_fp16 = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_262, y = transpose_261)[name = tensor<string, []>("qk_29_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1685_cast_fp16 = softmax(axis = var_1620, x = qk_29_cast_fp16)[name = tensor<string, []>("op_1685_cast_fp16")];
            tensor<bool, []> var_1687_transpose_x_0 = const()[name = tensor<string, []>("op_1687_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1687_transpose_y_0 = const()[name = tensor<string, []>("op_1687_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_263 = transpose(perm = var_1681, x = var_1680_cast_fp16)[name = tensor<string, []>("transpose_263")];
            tensor<fp16, [1, 20, 1500, 64]> var_1687_cast_fp16 = matmul(transpose_x = var_1687_transpose_x_0, transpose_y = var_1687_transpose_y_0, x = var_1685_cast_fp16, y = transpose_263)[name = tensor<string, []>("op_1687_cast_fp16")];
            tensor<int32, [4]> var_1688 = const()[name = tensor<string, []>("op_1688"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_260 = transpose(perm = var_1688, x = var_1687_cast_fp16)[name = tensor<string, []>("transpose_260")];
            tensor<fp16, [1, 1500, 1280]> x_179_cast_fp16 = reshape(shape = concat_14, x = transpose_260)[name = tensor<string, []>("x_179_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1693_to_fp16 = const()[name = tensor<string, []>("op_1693_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575448576)))];
            tensor<fp16, [1280]> var_1694_to_fp16 = const()[name = tensor<string, []>("op_1694_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578725440)))];
            tensor<fp16, [1, 1500, 1280]> linear_87_cast_fp16 = linear(bias = var_1694_to_fp16, weight = var_1693_to_fp16, x = x_179_cast_fp16)[name = tensor<string, []>("linear_87_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_181_cast_fp16 = add(x = x_175_cast_fp16, y = linear_87_cast_fp16)[name = tensor<string, []>("x_181_cast_fp16")];
            tensor<int32, [1]> var_1701_axes_0 = const()[name = tensor<string, []>("op_1701_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_14_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578728064)))];
            tensor<fp16, [1280]> blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578730688)))];
            tensor<fp16, [1, 1500, 1280]> var_1701_cast_fp16 = layer_norm(axes = var_1701_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = tensor<string, []>("op_1701_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1710_to_fp16 = const()[name = tensor<string, []>("op_1710_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578733312)))];
            tensor<fp16, [5120]> var_1711_to_fp16 = const()[name = tensor<string, []>("op_1711_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591840576)))];
            tensor<fp16, [1, 1500, 5120]> linear_88_cast_fp16 = linear(bias = var_1711_to_fp16, weight = var_1710_to_fp16, x = var_1701_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")];
            tensor<string, []> x_185_mode_0 = const()[name = tensor<string, []>("x_185_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_185_cast_fp16 = gelu(mode = x_185_mode_0, x = linear_88_cast_fp16)[name = tensor<string, []>("x_185_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1716_to_fp16 = const()[name = tensor<string, []>("op_1716_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591850880)))];
            tensor<fp16, [1280]> var_1717_to_fp16 = const()[name = tensor<string, []>("op_1717_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604958144)))];
            tensor<fp16, [1, 1500, 1280]> linear_89_cast_fp16 = linear(bias = var_1717_to_fp16, weight = var_1716_to_fp16, x = x_185_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_187_cast_fp16 = add(x = x_181_cast_fp16, y = linear_89_cast_fp16)[name = tensor<string, []>("x_187_cast_fp16")];
            tensor<int32, []> var_1727 = const()[name = tensor<string, []>("op_1727"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1744_axes_0 = const()[name = tensor<string, []>("op_1744_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_15_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604960768)))];
            tensor<fp16, [1280]> blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604963392)))];
            tensor<fp16, []> var_1733_to_fp16 = const()[name = tensor<string, []>("op_1733_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1744_cast_fp16 = layer_norm(axes = var_1744_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = tensor<string, []>("op_1744_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1755_to_fp16 = const()[name = tensor<string, []>("op_1755_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604966016)))];
            tensor<fp16, [1280]> var_1756_to_fp16 = const()[name = tensor<string, []>("op_1756_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608242880)))];
            tensor<fp16, [1, 1500, 1280]> linear_90_cast_fp16 = linear(bias = var_1756_to_fp16, weight = var_1755_to_fp16, x = var_1744_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1759_to_fp16 = const()[name = tensor<string, []>("op_1759_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608245504)))];
            tensor<fp16, [1, 1500, 1280]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1759_to_fp16, x = var_1744_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1763_to_fp16 = const()[name = tensor<string, []>("op_1763_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611522368)))];
            tensor<fp16, [1280]> var_1764_to_fp16 = const()[name = tensor<string, []>("op_1764_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614799232)))];
            tensor<fp16, [1, 1500, 1280]> linear_92_cast_fp16 = linear(bias = var_1764_to_fp16, weight = var_1763_to_fp16, x = var_1744_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")];
            tensor<int32, [4]> var_1772 = const()[name = tensor<string, []>("op_1772"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1773_cast_fp16 = reshape(shape = var_1772, x = linear_90_cast_fp16)[name = tensor<string, []>("op_1773_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_254_to_fp16 = const()[name = tensor<string, []>("const_254_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_63_cast_fp16 = mul(x = var_1773_cast_fp16, y = const_254_to_fp16)[name = tensor<string, []>("q_63_cast_fp16")];
            tensor<int32, [4]> var_1779 = const()[name = tensor<string, []>("op_1779"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1780_cast_fp16 = reshape(shape = var_1779, x = linear_91_cast_fp16)[name = tensor<string, []>("op_1780_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_255_to_fp16 = const()[name = tensor<string, []>("const_255_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_63_cast_fp16 = mul(x = var_1780_cast_fp16, y = const_255_to_fp16)[name = tensor<string, []>("k_63_cast_fp16")];
            tensor<int32, [4]> var_1786 = const()[name = tensor<string, []>("op_1786"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1787_cast_fp16 = reshape(shape = var_1786, x = linear_92_cast_fp16)[name = tensor<string, []>("op_1787_cast_fp16")];
            tensor<int32, [4]> var_1788 = const()[name = tensor<string, []>("op_1788"), 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_158_perm_0 = const()[name = tensor<string, []>("transpose_158_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_159_perm_0 = const()[name = tensor<string, []>("transpose_159_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_257 = transpose(perm = transpose_159_perm_0, x = k_63_cast_fp16)[name = tensor<string, []>("transpose_257")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_258 = transpose(perm = transpose_158_perm_0, x = q_63_cast_fp16)[name = tensor<string, []>("transpose_258")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_31_cast_fp16 = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_258, y = transpose_257)[name = tensor<string, []>("qk_31_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1792_cast_fp16 = softmax(axis = var_1727, x = qk_31_cast_fp16)[name = tensor<string, []>("op_1792_cast_fp16")];
            tensor<bool, []> var_1794_transpose_x_0 = const()[name = tensor<string, []>("op_1794_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1794_transpose_y_0 = const()[name = tensor<string, []>("op_1794_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_259 = transpose(perm = var_1788, x = var_1787_cast_fp16)[name = tensor<string, []>("transpose_259")];
            tensor<fp16, [1, 20, 1500, 64]> var_1794_cast_fp16 = matmul(transpose_x = var_1794_transpose_x_0, transpose_y = var_1794_transpose_y_0, x = var_1792_cast_fp16, y = transpose_259)[name = tensor<string, []>("op_1794_cast_fp16")];
            tensor<int32, [4]> var_1795 = const()[name = tensor<string, []>("op_1795"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_256 = transpose(perm = var_1795, x = var_1794_cast_fp16)[name = tensor<string, []>("transpose_256")];
            tensor<fp16, [1, 1500, 1280]> x_191_cast_fp16 = reshape(shape = concat_15, x = transpose_256)[name = tensor<string, []>("x_191_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1800_to_fp16 = const()[name = tensor<string, []>("op_1800_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614801856)))];
            tensor<fp16, [1280]> var_1801_to_fp16 = const()[name = tensor<string, []>("op_1801_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618078720)))];
            tensor<fp16, [1, 1500, 1280]> linear_93_cast_fp16 = linear(bias = var_1801_to_fp16, weight = var_1800_to_fp16, x = x_191_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_193_cast_fp16 = add(x = x_187_cast_fp16, y = linear_93_cast_fp16)[name = tensor<string, []>("x_193_cast_fp16")];
            tensor<int32, [1]> var_1808_axes_0 = const()[name = tensor<string, []>("op_1808_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_15_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618081344)))];
            tensor<fp16, [1280]> blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618083968)))];
            tensor<fp16, [1, 1500, 1280]> var_1808_cast_fp16 = layer_norm(axes = var_1808_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16)[name = tensor<string, []>("op_1808_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1817_to_fp16 = const()[name = tensor<string, []>("op_1817_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(618086592)))];
            tensor<fp16, [5120]> var_1818_to_fp16 = const()[name = tensor<string, []>("op_1818_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631193856)))];
            tensor<fp16, [1, 1500, 5120]> linear_94_cast_fp16 = linear(bias = var_1818_to_fp16, weight = var_1817_to_fp16, x = var_1808_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")];
            tensor<string, []> x_197_mode_0 = const()[name = tensor<string, []>("x_197_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_197_cast_fp16 = gelu(mode = x_197_mode_0, x = linear_94_cast_fp16)[name = tensor<string, []>("x_197_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1823_to_fp16 = const()[name = tensor<string, []>("op_1823_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631204160)))];
            tensor<fp16, [1280]> var_1824_to_fp16 = const()[name = tensor<string, []>("op_1824_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644311424)))];
            tensor<fp16, [1, 1500, 1280]> linear_95_cast_fp16 = linear(bias = var_1824_to_fp16, weight = var_1823_to_fp16, x = x_197_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_199_cast_fp16 = add(x = x_193_cast_fp16, y = linear_95_cast_fp16)[name = tensor<string, []>("x_199_cast_fp16")];
            tensor<int32, []> var_1834 = const()[name = tensor<string, []>("op_1834"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1851_axes_0 = const()[name = tensor<string, []>("op_1851_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_16_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644314048)))];
            tensor<fp16, [1280]> blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644316672)))];
            tensor<fp16, []> var_1840_to_fp16 = const()[name = tensor<string, []>("op_1840_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1851_cast_fp16 = layer_norm(axes = var_1851_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = tensor<string, []>("op_1851_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1862_to_fp16 = const()[name = tensor<string, []>("op_1862_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644319296)))];
            tensor<fp16, [1280]> var_1863_to_fp16 = const()[name = tensor<string, []>("op_1863_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(647596160)))];
            tensor<fp16, [1, 1500, 1280]> linear_96_cast_fp16 = linear(bias = var_1863_to_fp16, weight = var_1862_to_fp16, x = var_1851_cast_fp16)[name = tensor<string, []>("linear_96_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1866_to_fp16 = const()[name = tensor<string, []>("op_1866_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(647598784)))];
            tensor<fp16, [1, 1500, 1280]> linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1866_to_fp16, x = var_1851_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1870_to_fp16 = const()[name = tensor<string, []>("op_1870_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650875648)))];
            tensor<fp16, [1280]> var_1871_to_fp16 = const()[name = tensor<string, []>("op_1871_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(654152512)))];
            tensor<fp16, [1, 1500, 1280]> linear_98_cast_fp16 = linear(bias = var_1871_to_fp16, weight = var_1870_to_fp16, x = var_1851_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")];
            tensor<int32, [4]> var_1879 = const()[name = tensor<string, []>("op_1879"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1880_cast_fp16 = reshape(shape = var_1879, x = linear_96_cast_fp16)[name = tensor<string, []>("op_1880_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_256_to_fp16 = const()[name = tensor<string, []>("const_256_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_67_cast_fp16 = mul(x = var_1880_cast_fp16, y = const_256_to_fp16)[name = tensor<string, []>("q_67_cast_fp16")];
            tensor<int32, [4]> var_1886 = const()[name = tensor<string, []>("op_1886"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1887_cast_fp16 = reshape(shape = var_1886, x = linear_97_cast_fp16)[name = tensor<string, []>("op_1887_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_257_to_fp16 = const()[name = tensor<string, []>("const_257_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_67_cast_fp16 = mul(x = var_1887_cast_fp16, y = const_257_to_fp16)[name = tensor<string, []>("k_67_cast_fp16")];
            tensor<int32, [4]> var_1893 = const()[name = tensor<string, []>("op_1893"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1894_cast_fp16 = reshape(shape = var_1893, x = linear_98_cast_fp16)[name = tensor<string, []>("op_1894_cast_fp16")];
            tensor<int32, [4]> var_1895 = const()[name = tensor<string, []>("op_1895"), 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_160_perm_0 = const()[name = tensor<string, []>("transpose_160_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_161_perm_0 = const()[name = tensor<string, []>("transpose_161_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_253 = transpose(perm = transpose_161_perm_0, x = k_67_cast_fp16)[name = tensor<string, []>("transpose_253")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_254 = transpose(perm = transpose_160_perm_0, x = q_67_cast_fp16)[name = tensor<string, []>("transpose_254")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_33_cast_fp16 = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_254, y = transpose_253)[name = tensor<string, []>("qk_33_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_1899_cast_fp16 = softmax(axis = var_1834, x = qk_33_cast_fp16)[name = tensor<string, []>("op_1899_cast_fp16")];
            tensor<bool, []> var_1901_transpose_x_0 = const()[name = tensor<string, []>("op_1901_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1901_transpose_y_0 = const()[name = tensor<string, []>("op_1901_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_255 = transpose(perm = var_1895, x = var_1894_cast_fp16)[name = tensor<string, []>("transpose_255")];
            tensor<fp16, [1, 20, 1500, 64]> var_1901_cast_fp16 = matmul(transpose_x = var_1901_transpose_x_0, transpose_y = var_1901_transpose_y_0, x = var_1899_cast_fp16, y = transpose_255)[name = tensor<string, []>("op_1901_cast_fp16")];
            tensor<int32, [4]> var_1902 = const()[name = tensor<string, []>("op_1902"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_252 = transpose(perm = var_1902, x = var_1901_cast_fp16)[name = tensor<string, []>("transpose_252")];
            tensor<fp16, [1, 1500, 1280]> x_203_cast_fp16 = reshape(shape = concat_16, x = transpose_252)[name = tensor<string, []>("x_203_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1907_to_fp16 = const()[name = tensor<string, []>("op_1907_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(654155136)))];
            tensor<fp16, [1280]> var_1908_to_fp16 = const()[name = tensor<string, []>("op_1908_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657432000)))];
            tensor<fp16, [1, 1500, 1280]> linear_99_cast_fp16 = linear(bias = var_1908_to_fp16, weight = var_1907_to_fp16, x = x_203_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_205_cast_fp16 = add(x = x_199_cast_fp16, y = linear_99_cast_fp16)[name = tensor<string, []>("x_205_cast_fp16")];
            tensor<int32, [1]> var_1915_axes_0 = const()[name = tensor<string, []>("op_1915_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_16_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657434624)))];
            tensor<fp16, [1280]> blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657437248)))];
            tensor<fp16, [1, 1500, 1280]> var_1915_cast_fp16 = layer_norm(axes = var_1915_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = tensor<string, []>("op_1915_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_1924_to_fp16 = const()[name = tensor<string, []>("op_1924_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657439872)))];
            tensor<fp16, [5120]> var_1925_to_fp16 = const()[name = tensor<string, []>("op_1925_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670547136)))];
            tensor<fp16, [1, 1500, 5120]> linear_100_cast_fp16 = linear(bias = var_1925_to_fp16, weight = var_1924_to_fp16, x = var_1915_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")];
            tensor<string, []> x_209_mode_0 = const()[name = tensor<string, []>("x_209_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_209_cast_fp16 = gelu(mode = x_209_mode_0, x = linear_100_cast_fp16)[name = tensor<string, []>("x_209_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_1930_to_fp16 = const()[name = tensor<string, []>("op_1930_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670557440)))];
            tensor<fp16, [1280]> var_1931_to_fp16 = const()[name = tensor<string, []>("op_1931_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683664704)))];
            tensor<fp16, [1, 1500, 1280]> linear_101_cast_fp16 = linear(bias = var_1931_to_fp16, weight = var_1930_to_fp16, x = x_209_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_211_cast_fp16 = add(x = x_205_cast_fp16, y = linear_101_cast_fp16)[name = tensor<string, []>("x_211_cast_fp16")];
            tensor<int32, []> var_1941 = const()[name = tensor<string, []>("op_1941"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1958_axes_0 = const()[name = tensor<string, []>("op_1958_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_17_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683667328)))];
            tensor<fp16, [1280]> blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683669952)))];
            tensor<fp16, []> var_1947_to_fp16 = const()[name = tensor<string, []>("op_1947_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1958_cast_fp16 = layer_norm(axes = var_1958_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = tensor<string, []>("op_1958_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1969_to_fp16 = const()[name = tensor<string, []>("op_1969_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683672576)))];
            tensor<fp16, [1280]> var_1970_to_fp16 = const()[name = tensor<string, []>("op_1970_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686949440)))];
            tensor<fp16, [1, 1500, 1280]> linear_102_cast_fp16 = linear(bias = var_1970_to_fp16, weight = var_1969_to_fp16, x = var_1958_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1973_to_fp16 = const()[name = tensor<string, []>("op_1973_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686952064)))];
            tensor<fp16, [1, 1500, 1280]> linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1973_to_fp16, x = var_1958_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1977_to_fp16 = const()[name = tensor<string, []>("op_1977_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(690228928)))];
            tensor<fp16, [1280]> var_1978_to_fp16 = const()[name = tensor<string, []>("op_1978_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(693505792)))];
            tensor<fp16, [1, 1500, 1280]> linear_104_cast_fp16 = linear(bias = var_1978_to_fp16, weight = var_1977_to_fp16, x = var_1958_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")];
            tensor<int32, [4]> var_1986 = const()[name = tensor<string, []>("op_1986"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1987_cast_fp16 = reshape(shape = var_1986, x = linear_102_cast_fp16)[name = tensor<string, []>("op_1987_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_258_to_fp16 = const()[name = tensor<string, []>("const_258_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_71_cast_fp16 = mul(x = var_1987_cast_fp16, y = const_258_to_fp16)[name = tensor<string, []>("q_71_cast_fp16")];
            tensor<int32, [4]> var_1993 = const()[name = tensor<string, []>("op_1993"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1994_cast_fp16 = reshape(shape = var_1993, x = linear_103_cast_fp16)[name = tensor<string, []>("op_1994_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_259_to_fp16 = const()[name = tensor<string, []>("const_259_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_71_cast_fp16 = mul(x = var_1994_cast_fp16, y = const_259_to_fp16)[name = tensor<string, []>("k_71_cast_fp16")];
            tensor<int32, [4]> var_2000 = const()[name = tensor<string, []>("op_2000"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2001_cast_fp16 = reshape(shape = var_2000, x = linear_104_cast_fp16)[name = tensor<string, []>("op_2001_cast_fp16")];
            tensor<int32, [4]> var_2002 = const()[name = tensor<string, []>("op_2002"), 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_162_perm_0 = const()[name = tensor<string, []>("transpose_162_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_163_perm_0 = const()[name = tensor<string, []>("transpose_163_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_249 = transpose(perm = transpose_163_perm_0, x = k_71_cast_fp16)[name = tensor<string, []>("transpose_249")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_250 = transpose(perm = transpose_162_perm_0, x = q_71_cast_fp16)[name = tensor<string, []>("transpose_250")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_35_cast_fp16 = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_250, y = transpose_249)[name = tensor<string, []>("qk_35_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2006_cast_fp16 = softmax(axis = var_1941, x = qk_35_cast_fp16)[name = tensor<string, []>("op_2006_cast_fp16")];
            tensor<bool, []> var_2008_transpose_x_0 = const()[name = tensor<string, []>("op_2008_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2008_transpose_y_0 = const()[name = tensor<string, []>("op_2008_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_251 = transpose(perm = var_2002, x = var_2001_cast_fp16)[name = tensor<string, []>("transpose_251")];
            tensor<fp16, [1, 20, 1500, 64]> var_2008_cast_fp16 = matmul(transpose_x = var_2008_transpose_x_0, transpose_y = var_2008_transpose_y_0, x = var_2006_cast_fp16, y = transpose_251)[name = tensor<string, []>("op_2008_cast_fp16")];
            tensor<int32, [4]> var_2009 = const()[name = tensor<string, []>("op_2009"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_248 = transpose(perm = var_2009, x = var_2008_cast_fp16)[name = tensor<string, []>("transpose_248")];
            tensor<fp16, [1, 1500, 1280]> x_215_cast_fp16 = reshape(shape = concat_17, x = transpose_248)[name = tensor<string, []>("x_215_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2014_to_fp16 = const()[name = tensor<string, []>("op_2014_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(693508416)))];
            tensor<fp16, [1280]> var_2015_to_fp16 = const()[name = tensor<string, []>("op_2015_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696785280)))];
            tensor<fp16, [1, 1500, 1280]> linear_105_cast_fp16 = linear(bias = var_2015_to_fp16, weight = var_2014_to_fp16, x = x_215_cast_fp16)[name = tensor<string, []>("linear_105_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_217_cast_fp16 = add(x = x_211_cast_fp16, y = linear_105_cast_fp16)[name = tensor<string, []>("x_217_cast_fp16")];
            tensor<int32, [1]> var_2022_axes_0 = const()[name = tensor<string, []>("op_2022_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_17_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696787904)))];
            tensor<fp16, [1280]> blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696790528)))];
            tensor<fp16, [1, 1500, 1280]> var_2022_cast_fp16 = layer_norm(axes = var_2022_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = tensor<string, []>("op_2022_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2031_to_fp16 = const()[name = tensor<string, []>("op_2031_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696793152)))];
            tensor<fp16, [5120]> var_2032_to_fp16 = const()[name = tensor<string, []>("op_2032_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709900416)))];
            tensor<fp16, [1, 1500, 5120]> linear_106_cast_fp16 = linear(bias = var_2032_to_fp16, weight = var_2031_to_fp16, x = var_2022_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")];
            tensor<string, []> x_221_mode_0 = const()[name = tensor<string, []>("x_221_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_221_cast_fp16 = gelu(mode = x_221_mode_0, x = linear_106_cast_fp16)[name = tensor<string, []>("x_221_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2037_to_fp16 = const()[name = tensor<string, []>("op_2037_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709910720)))];
            tensor<fp16, [1280]> var_2038_to_fp16 = const()[name = tensor<string, []>("op_2038_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723017984)))];
            tensor<fp16, [1, 1500, 1280]> linear_107_cast_fp16 = linear(bias = var_2038_to_fp16, weight = var_2037_to_fp16, x = x_221_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_223_cast_fp16 = add(x = x_217_cast_fp16, y = linear_107_cast_fp16)[name = tensor<string, []>("x_223_cast_fp16")];
            tensor<int32, []> var_2048 = const()[name = tensor<string, []>("op_2048"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2065_axes_0 = const()[name = tensor<string, []>("op_2065_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_18_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723020608)))];
            tensor<fp16, [1280]> blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723023232)))];
            tensor<fp16, []> var_2054_to_fp16 = const()[name = tensor<string, []>("op_2054_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2065_cast_fp16 = layer_norm(axes = var_2065_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = tensor<string, []>("op_2065_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2076_to_fp16 = const()[name = tensor<string, []>("op_2076_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723025856)))];
            tensor<fp16, [1280]> var_2077_to_fp16 = const()[name = tensor<string, []>("op_2077_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(726302720)))];
            tensor<fp16, [1, 1500, 1280]> linear_108_cast_fp16 = linear(bias = var_2077_to_fp16, weight = var_2076_to_fp16, x = var_2065_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2080_to_fp16 = const()[name = tensor<string, []>("op_2080_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(726305344)))];
            tensor<fp16, [1, 1500, 1280]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2080_to_fp16, x = var_2065_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2084_to_fp16 = const()[name = tensor<string, []>("op_2084_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(729582208)))];
            tensor<fp16, [1280]> var_2085_to_fp16 = const()[name = tensor<string, []>("op_2085_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732859072)))];
            tensor<fp16, [1, 1500, 1280]> linear_110_cast_fp16 = linear(bias = var_2085_to_fp16, weight = var_2084_to_fp16, x = var_2065_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")];
            tensor<int32, [4]> var_2093 = const()[name = tensor<string, []>("op_2093"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2094_cast_fp16 = reshape(shape = var_2093, x = linear_108_cast_fp16)[name = tensor<string, []>("op_2094_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_260_to_fp16 = const()[name = tensor<string, []>("const_260_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_75_cast_fp16 = mul(x = var_2094_cast_fp16, y = const_260_to_fp16)[name = tensor<string, []>("q_75_cast_fp16")];
            tensor<int32, [4]> var_2100 = const()[name = tensor<string, []>("op_2100"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2101_cast_fp16 = reshape(shape = var_2100, x = linear_109_cast_fp16)[name = tensor<string, []>("op_2101_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_261_to_fp16 = const()[name = tensor<string, []>("const_261_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_75_cast_fp16 = mul(x = var_2101_cast_fp16, y = const_261_to_fp16)[name = tensor<string, []>("k_75_cast_fp16")];
            tensor<int32, [4]> var_2107 = const()[name = tensor<string, []>("op_2107"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2108_cast_fp16 = reshape(shape = var_2107, x = linear_110_cast_fp16)[name = tensor<string, []>("op_2108_cast_fp16")];
            tensor<int32, [4]> var_2109 = const()[name = tensor<string, []>("op_2109"), 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_164_perm_0 = const()[name = tensor<string, []>("transpose_164_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_165_perm_0 = const()[name = tensor<string, []>("transpose_165_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_245 = transpose(perm = transpose_165_perm_0, x = k_75_cast_fp16)[name = tensor<string, []>("transpose_245")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_246 = transpose(perm = transpose_164_perm_0, x = q_75_cast_fp16)[name = tensor<string, []>("transpose_246")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_37_cast_fp16 = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_246, y = transpose_245)[name = tensor<string, []>("qk_37_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2113_cast_fp16 = softmax(axis = var_2048, x = qk_37_cast_fp16)[name = tensor<string, []>("op_2113_cast_fp16")];
            tensor<bool, []> var_2115_transpose_x_0 = const()[name = tensor<string, []>("op_2115_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2115_transpose_y_0 = const()[name = tensor<string, []>("op_2115_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_247 = transpose(perm = var_2109, x = var_2108_cast_fp16)[name = tensor<string, []>("transpose_247")];
            tensor<fp16, [1, 20, 1500, 64]> var_2115_cast_fp16 = matmul(transpose_x = var_2115_transpose_x_0, transpose_y = var_2115_transpose_y_0, x = var_2113_cast_fp16, y = transpose_247)[name = tensor<string, []>("op_2115_cast_fp16")];
            tensor<int32, [4]> var_2116 = const()[name = tensor<string, []>("op_2116"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_244 = transpose(perm = var_2116, x = var_2115_cast_fp16)[name = tensor<string, []>("transpose_244")];
            tensor<fp16, [1, 1500, 1280]> x_227_cast_fp16 = reshape(shape = concat_18, x = transpose_244)[name = tensor<string, []>("x_227_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2121_to_fp16 = const()[name = tensor<string, []>("op_2121_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732861696)))];
            tensor<fp16, [1280]> var_2122_to_fp16 = const()[name = tensor<string, []>("op_2122_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736138560)))];
            tensor<fp16, [1, 1500, 1280]> linear_111_cast_fp16 = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = x_227_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_229_cast_fp16 = add(x = x_223_cast_fp16, y = linear_111_cast_fp16)[name = tensor<string, []>("x_229_cast_fp16")];
            tensor<int32, [1]> var_2129_axes_0 = const()[name = tensor<string, []>("op_2129_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_18_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736141184)))];
            tensor<fp16, [1280]> blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736143808)))];
            tensor<fp16, [1, 1500, 1280]> var_2129_cast_fp16 = layer_norm(axes = var_2129_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = tensor<string, []>("op_2129_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2138_to_fp16 = const()[name = tensor<string, []>("op_2138_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736146432)))];
            tensor<fp16, [5120]> var_2139_to_fp16 = const()[name = tensor<string, []>("op_2139_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749253696)))];
            tensor<fp16, [1, 1500, 5120]> linear_112_cast_fp16 = linear(bias = var_2139_to_fp16, weight = var_2138_to_fp16, x = var_2129_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")];
            tensor<string, []> x_233_mode_0 = const()[name = tensor<string, []>("x_233_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_233_cast_fp16 = gelu(mode = x_233_mode_0, x = linear_112_cast_fp16)[name = tensor<string, []>("x_233_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2144_to_fp16 = const()[name = tensor<string, []>("op_2144_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749264000)))];
            tensor<fp16, [1280]> var_2145_to_fp16 = const()[name = tensor<string, []>("op_2145_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762371264)))];
            tensor<fp16, [1, 1500, 1280]> linear_113_cast_fp16 = linear(bias = var_2145_to_fp16, weight = var_2144_to_fp16, x = x_233_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_235_cast_fp16 = add(x = x_229_cast_fp16, y = linear_113_cast_fp16)[name = tensor<string, []>("x_235_cast_fp16")];
            tensor<int32, []> var_2155 = const()[name = tensor<string, []>("op_2155"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2172_axes_0 = const()[name = tensor<string, []>("op_2172_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_19_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762373888)))];
            tensor<fp16, [1280]> blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762376512)))];
            tensor<fp16, []> var_2161_to_fp16 = const()[name = tensor<string, []>("op_2161_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2172_cast_fp16 = layer_norm(axes = var_2172_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = tensor<string, []>("op_2172_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2183_to_fp16 = const()[name = tensor<string, []>("op_2183_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762379136)))];
            tensor<fp16, [1280]> var_2184_to_fp16 = const()[name = tensor<string, []>("op_2184_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765656000)))];
            tensor<fp16, [1, 1500, 1280]> linear_114_cast_fp16 = linear(bias = var_2184_to_fp16, weight = var_2183_to_fp16, x = var_2172_cast_fp16)[name = tensor<string, []>("linear_114_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2187_to_fp16 = const()[name = tensor<string, []>("op_2187_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765658624)))];
            tensor<fp16, [1, 1500, 1280]> linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2187_to_fp16, x = var_2172_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2191_to_fp16 = const()[name = tensor<string, []>("op_2191_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(768935488)))];
            tensor<fp16, [1280]> var_2192_to_fp16 = const()[name = tensor<string, []>("op_2192_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772212352)))];
            tensor<fp16, [1, 1500, 1280]> linear_116_cast_fp16 = linear(bias = var_2192_to_fp16, weight = var_2191_to_fp16, x = var_2172_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")];
            tensor<int32, [4]> var_2200 = const()[name = tensor<string, []>("op_2200"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2201_cast_fp16 = reshape(shape = var_2200, x = linear_114_cast_fp16)[name = tensor<string, []>("op_2201_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_262_to_fp16 = const()[name = tensor<string, []>("const_262_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_79_cast_fp16 = mul(x = var_2201_cast_fp16, y = const_262_to_fp16)[name = tensor<string, []>("q_79_cast_fp16")];
            tensor<int32, [4]> var_2207 = const()[name = tensor<string, []>("op_2207"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2208_cast_fp16 = reshape(shape = var_2207, x = linear_115_cast_fp16)[name = tensor<string, []>("op_2208_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_263_to_fp16 = const()[name = tensor<string, []>("const_263_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_79_cast_fp16 = mul(x = var_2208_cast_fp16, y = const_263_to_fp16)[name = tensor<string, []>("k_79_cast_fp16")];
            tensor<int32, [4]> var_2214 = const()[name = tensor<string, []>("op_2214"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2215_cast_fp16 = reshape(shape = var_2214, x = linear_116_cast_fp16)[name = tensor<string, []>("op_2215_cast_fp16")];
            tensor<int32, [4]> var_2216 = const()[name = tensor<string, []>("op_2216"), 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_166_perm_0 = const()[name = tensor<string, []>("transpose_166_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_167_perm_0 = const()[name = tensor<string, []>("transpose_167_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_241 = transpose(perm = transpose_167_perm_0, x = k_79_cast_fp16)[name = tensor<string, []>("transpose_241")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_242 = transpose(perm = transpose_166_perm_0, x = q_79_cast_fp16)[name = tensor<string, []>("transpose_242")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_39_cast_fp16 = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_242, y = transpose_241)[name = tensor<string, []>("qk_39_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2220_cast_fp16 = softmax(axis = var_2155, x = qk_39_cast_fp16)[name = tensor<string, []>("op_2220_cast_fp16")];
            tensor<bool, []> var_2222_transpose_x_0 = const()[name = tensor<string, []>("op_2222_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2222_transpose_y_0 = const()[name = tensor<string, []>("op_2222_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_243 = transpose(perm = var_2216, x = var_2215_cast_fp16)[name = tensor<string, []>("transpose_243")];
            tensor<fp16, [1, 20, 1500, 64]> var_2222_cast_fp16 = matmul(transpose_x = var_2222_transpose_x_0, transpose_y = var_2222_transpose_y_0, x = var_2220_cast_fp16, y = transpose_243)[name = tensor<string, []>("op_2222_cast_fp16")];
            tensor<int32, [4]> var_2223 = const()[name = tensor<string, []>("op_2223"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_240 = transpose(perm = var_2223, x = var_2222_cast_fp16)[name = tensor<string, []>("transpose_240")];
            tensor<fp16, [1, 1500, 1280]> x_239_cast_fp16 = reshape(shape = concat_19, x = transpose_240)[name = tensor<string, []>("x_239_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2228_to_fp16 = const()[name = tensor<string, []>("op_2228_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772214976)))];
            tensor<fp16, [1280]> var_2229_to_fp16 = const()[name = tensor<string, []>("op_2229_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775491840)))];
            tensor<fp16, [1, 1500, 1280]> linear_117_cast_fp16 = linear(bias = var_2229_to_fp16, weight = var_2228_to_fp16, x = x_239_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_241_cast_fp16 = add(x = x_235_cast_fp16, y = linear_117_cast_fp16)[name = tensor<string, []>("x_241_cast_fp16")];
            tensor<int32, [1]> var_2236_axes_0 = const()[name = tensor<string, []>("op_2236_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_19_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775494464)))];
            tensor<fp16, [1280]> blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775497088)))];
            tensor<fp16, [1, 1500, 1280]> var_2236_cast_fp16 = layer_norm(axes = var_2236_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = tensor<string, []>("op_2236_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2245_to_fp16 = const()[name = tensor<string, []>("op_2245_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775499712)))];
            tensor<fp16, [5120]> var_2246_to_fp16 = const()[name = tensor<string, []>("op_2246_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(788606976)))];
            tensor<fp16, [1, 1500, 5120]> linear_118_cast_fp16 = linear(bias = var_2246_to_fp16, weight = var_2245_to_fp16, x = var_2236_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")];
            tensor<string, []> x_245_mode_0 = const()[name = tensor<string, []>("x_245_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_245_cast_fp16 = gelu(mode = x_245_mode_0, x = linear_118_cast_fp16)[name = tensor<string, []>("x_245_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2251_to_fp16 = const()[name = tensor<string, []>("op_2251_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(788617280)))];
            tensor<fp16, [1280]> var_2252_to_fp16 = const()[name = tensor<string, []>("op_2252_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801724544)))];
            tensor<fp16, [1, 1500, 1280]> linear_119_cast_fp16 = linear(bias = var_2252_to_fp16, weight = var_2251_to_fp16, x = x_245_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_247_cast_fp16 = add(x = x_241_cast_fp16, y = linear_119_cast_fp16)[name = tensor<string, []>("x_247_cast_fp16")];
            tensor<int32, []> var_2262 = const()[name = tensor<string, []>("op_2262"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2279_axes_0 = const()[name = tensor<string, []>("op_2279_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_20_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801727168)))];
            tensor<fp16, [1280]> blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801729792)))];
            tensor<fp16, []> var_2268_to_fp16 = const()[name = tensor<string, []>("op_2268_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2279_cast_fp16 = layer_norm(axes = var_2279_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = tensor<string, []>("op_2279_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2290_to_fp16 = const()[name = tensor<string, []>("op_2290_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801732416)))];
            tensor<fp16, [1280]> var_2291_to_fp16 = const()[name = tensor<string, []>("op_2291_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805009280)))];
            tensor<fp16, [1, 1500, 1280]> linear_120_cast_fp16 = linear(bias = var_2291_to_fp16, weight = var_2290_to_fp16, x = var_2279_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2294_to_fp16 = const()[name = tensor<string, []>("op_2294_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805011904)))];
            tensor<fp16, [1, 1500, 1280]> linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2294_to_fp16, x = var_2279_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2298_to_fp16 = const()[name = tensor<string, []>("op_2298_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808288768)))];
            tensor<fp16, [1280]> var_2299_to_fp16 = const()[name = tensor<string, []>("op_2299_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811565632)))];
            tensor<fp16, [1, 1500, 1280]> linear_122_cast_fp16 = linear(bias = var_2299_to_fp16, weight = var_2298_to_fp16, x = var_2279_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")];
            tensor<int32, [4]> var_2307 = const()[name = tensor<string, []>("op_2307"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2308_cast_fp16 = reshape(shape = var_2307, x = linear_120_cast_fp16)[name = tensor<string, []>("op_2308_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_264_to_fp16 = const()[name = tensor<string, []>("const_264_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_83_cast_fp16 = mul(x = var_2308_cast_fp16, y = const_264_to_fp16)[name = tensor<string, []>("q_83_cast_fp16")];
            tensor<int32, [4]> var_2314 = const()[name = tensor<string, []>("op_2314"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2315_cast_fp16 = reshape(shape = var_2314, x = linear_121_cast_fp16)[name = tensor<string, []>("op_2315_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_265_to_fp16 = const()[name = tensor<string, []>("const_265_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_83_cast_fp16 = mul(x = var_2315_cast_fp16, y = const_265_to_fp16)[name = tensor<string, []>("k_83_cast_fp16")];
            tensor<int32, [4]> var_2321 = const()[name = tensor<string, []>("op_2321"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2322_cast_fp16 = reshape(shape = var_2321, x = linear_122_cast_fp16)[name = tensor<string, []>("op_2322_cast_fp16")];
            tensor<int32, [4]> var_2323 = const()[name = tensor<string, []>("op_2323"), 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_168_perm_0 = const()[name = tensor<string, []>("transpose_168_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_169_perm_0 = const()[name = tensor<string, []>("transpose_169_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_237 = transpose(perm = transpose_169_perm_0, x = k_83_cast_fp16)[name = tensor<string, []>("transpose_237")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_238 = transpose(perm = transpose_168_perm_0, x = q_83_cast_fp16)[name = tensor<string, []>("transpose_238")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_41_cast_fp16 = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_238, y = transpose_237)[name = tensor<string, []>("qk_41_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2327_cast_fp16 = softmax(axis = var_2262, x = qk_41_cast_fp16)[name = tensor<string, []>("op_2327_cast_fp16")];
            tensor<bool, []> var_2329_transpose_x_0 = const()[name = tensor<string, []>("op_2329_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2329_transpose_y_0 = const()[name = tensor<string, []>("op_2329_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_239 = transpose(perm = var_2323, x = var_2322_cast_fp16)[name = tensor<string, []>("transpose_239")];
            tensor<fp16, [1, 20, 1500, 64]> var_2329_cast_fp16 = matmul(transpose_x = var_2329_transpose_x_0, transpose_y = var_2329_transpose_y_0, x = var_2327_cast_fp16, y = transpose_239)[name = tensor<string, []>("op_2329_cast_fp16")];
            tensor<int32, [4]> var_2330 = const()[name = tensor<string, []>("op_2330"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_236 = transpose(perm = var_2330, x = var_2329_cast_fp16)[name = tensor<string, []>("transpose_236")];
            tensor<fp16, [1, 1500, 1280]> x_251_cast_fp16 = reshape(shape = concat_20, x = transpose_236)[name = tensor<string, []>("x_251_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2335_to_fp16 = const()[name = tensor<string, []>("op_2335_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811568256)))];
            tensor<fp16, [1280]> var_2336_to_fp16 = const()[name = tensor<string, []>("op_2336_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814845120)))];
            tensor<fp16, [1, 1500, 1280]> linear_123_cast_fp16 = linear(bias = var_2336_to_fp16, weight = var_2335_to_fp16, x = x_251_cast_fp16)[name = tensor<string, []>("linear_123_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_253_cast_fp16 = add(x = x_247_cast_fp16, y = linear_123_cast_fp16)[name = tensor<string, []>("x_253_cast_fp16")];
            tensor<int32, [1]> var_2343_axes_0 = const()[name = tensor<string, []>("op_2343_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_20_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814847744)))];
            tensor<fp16, [1280]> blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814850368)))];
            tensor<fp16, [1, 1500, 1280]> var_2343_cast_fp16 = layer_norm(axes = var_2343_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = tensor<string, []>("op_2343_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2352_to_fp16 = const()[name = tensor<string, []>("op_2352_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814852992)))];
            tensor<fp16, [5120]> var_2353_to_fp16 = const()[name = tensor<string, []>("op_2353_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(827960256)))];
            tensor<fp16, [1, 1500, 5120]> linear_124_cast_fp16 = linear(bias = var_2353_to_fp16, weight = var_2352_to_fp16, x = var_2343_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")];
            tensor<string, []> x_257_mode_0 = const()[name = tensor<string, []>("x_257_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_257_cast_fp16 = gelu(mode = x_257_mode_0, x = linear_124_cast_fp16)[name = tensor<string, []>("x_257_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2358_to_fp16 = const()[name = tensor<string, []>("op_2358_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(827970560)))];
            tensor<fp16, [1280]> var_2359_to_fp16 = const()[name = tensor<string, []>("op_2359_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841077824)))];
            tensor<fp16, [1, 1500, 1280]> linear_125_cast_fp16 = linear(bias = var_2359_to_fp16, weight = var_2358_to_fp16, x = x_257_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_259_cast_fp16 = add(x = x_253_cast_fp16, y = linear_125_cast_fp16)[name = tensor<string, []>("x_259_cast_fp16")];
            tensor<int32, []> var_2369 = const()[name = tensor<string, []>("op_2369"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2386_axes_0 = const()[name = tensor<string, []>("op_2386_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_21_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841080448)))];
            tensor<fp16, [1280]> blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841083072)))];
            tensor<fp16, []> var_2375_to_fp16 = const()[name = tensor<string, []>("op_2375_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2386_cast_fp16 = layer_norm(axes = var_2386_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = tensor<string, []>("op_2386_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2397_to_fp16 = const()[name = tensor<string, []>("op_2397_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841085696)))];
            tensor<fp16, [1280]> var_2398_to_fp16 = const()[name = tensor<string, []>("op_2398_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844362560)))];
            tensor<fp16, [1, 1500, 1280]> linear_126_cast_fp16 = linear(bias = var_2398_to_fp16, weight = var_2397_to_fp16, x = var_2386_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2401_to_fp16 = const()[name = tensor<string, []>("op_2401_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844365184)))];
            tensor<fp16, [1, 1500, 1280]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2401_to_fp16, x = var_2386_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2405_to_fp16 = const()[name = tensor<string, []>("op_2405_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(847642048)))];
            tensor<fp16, [1280]> var_2406_to_fp16 = const()[name = tensor<string, []>("op_2406_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(850918912)))];
            tensor<fp16, [1, 1500, 1280]> linear_128_cast_fp16 = linear(bias = var_2406_to_fp16, weight = var_2405_to_fp16, x = var_2386_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")];
            tensor<int32, [4]> var_2414 = const()[name = tensor<string, []>("op_2414"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2415_cast_fp16 = reshape(shape = var_2414, x = linear_126_cast_fp16)[name = tensor<string, []>("op_2415_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_266_to_fp16 = const()[name = tensor<string, []>("const_266_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_87_cast_fp16 = mul(x = var_2415_cast_fp16, y = const_266_to_fp16)[name = tensor<string, []>("q_87_cast_fp16")];
            tensor<int32, [4]> var_2421 = const()[name = tensor<string, []>("op_2421"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2422_cast_fp16 = reshape(shape = var_2421, x = linear_127_cast_fp16)[name = tensor<string, []>("op_2422_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_267_to_fp16 = const()[name = tensor<string, []>("const_267_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_87_cast_fp16 = mul(x = var_2422_cast_fp16, y = const_267_to_fp16)[name = tensor<string, []>("k_87_cast_fp16")];
            tensor<int32, [4]> var_2428 = const()[name = tensor<string, []>("op_2428"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2429_cast_fp16 = reshape(shape = var_2428, x = linear_128_cast_fp16)[name = tensor<string, []>("op_2429_cast_fp16")];
            tensor<int32, [4]> var_2430 = const()[name = tensor<string, []>("op_2430"), 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_170_perm_0 = const()[name = tensor<string, []>("transpose_170_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_171_perm_0 = const()[name = tensor<string, []>("transpose_171_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_233 = transpose(perm = transpose_171_perm_0, x = k_87_cast_fp16)[name = tensor<string, []>("transpose_233")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_234 = transpose(perm = transpose_170_perm_0, x = q_87_cast_fp16)[name = tensor<string, []>("transpose_234")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_43_cast_fp16 = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_234, y = transpose_233)[name = tensor<string, []>("qk_43_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2434_cast_fp16 = softmax(axis = var_2369, x = qk_43_cast_fp16)[name = tensor<string, []>("op_2434_cast_fp16")];
            tensor<bool, []> var_2436_transpose_x_0 = const()[name = tensor<string, []>("op_2436_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2436_transpose_y_0 = const()[name = tensor<string, []>("op_2436_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_235 = transpose(perm = var_2430, x = var_2429_cast_fp16)[name = tensor<string, []>("transpose_235")];
            tensor<fp16, [1, 20, 1500, 64]> var_2436_cast_fp16 = matmul(transpose_x = var_2436_transpose_x_0, transpose_y = var_2436_transpose_y_0, x = var_2434_cast_fp16, y = transpose_235)[name = tensor<string, []>("op_2436_cast_fp16")];
            tensor<int32, [4]> var_2437 = const()[name = tensor<string, []>("op_2437"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_232 = transpose(perm = var_2437, x = var_2436_cast_fp16)[name = tensor<string, []>("transpose_232")];
            tensor<fp16, [1, 1500, 1280]> x_263_cast_fp16 = reshape(shape = concat_21, x = transpose_232)[name = tensor<string, []>("x_263_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2442_to_fp16 = const()[name = tensor<string, []>("op_2442_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(850921536)))];
            tensor<fp16, [1280]> var_2443_to_fp16 = const()[name = tensor<string, []>("op_2443_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854198400)))];
            tensor<fp16, [1, 1500, 1280]> linear_129_cast_fp16 = linear(bias = var_2443_to_fp16, weight = var_2442_to_fp16, x = x_263_cast_fp16)[name = tensor<string, []>("linear_129_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_265_cast_fp16 = add(x = x_259_cast_fp16, y = linear_129_cast_fp16)[name = tensor<string, []>("x_265_cast_fp16")];
            tensor<int32, [1]> var_2450_axes_0 = const()[name = tensor<string, []>("op_2450_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_21_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854201024)))];
            tensor<fp16, [1280]> blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854203648)))];
            tensor<fp16, [1, 1500, 1280]> var_2450_cast_fp16 = layer_norm(axes = var_2450_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = tensor<string, []>("op_2450_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2459_to_fp16 = const()[name = tensor<string, []>("op_2459_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854206272)))];
            tensor<fp16, [5120]> var_2460_to_fp16 = const()[name = tensor<string, []>("op_2460_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(867313536)))];
            tensor<fp16, [1, 1500, 5120]> linear_130_cast_fp16 = linear(bias = var_2460_to_fp16, weight = var_2459_to_fp16, x = var_2450_cast_fp16)[name = tensor<string, []>("linear_130_cast_fp16")];
            tensor<string, []> x_269_mode_0 = const()[name = tensor<string, []>("x_269_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_269_cast_fp16 = gelu(mode = x_269_mode_0, x = linear_130_cast_fp16)[name = tensor<string, []>("x_269_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2465_to_fp16 = const()[name = tensor<string, []>("op_2465_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(867323840)))];
            tensor<fp16, [1280]> var_2466_to_fp16 = const()[name = tensor<string, []>("op_2466_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880431104)))];
            tensor<fp16, [1, 1500, 1280]> linear_131_cast_fp16 = linear(bias = var_2466_to_fp16, weight = var_2465_to_fp16, x = x_269_cast_fp16)[name = tensor<string, []>("linear_131_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_271_cast_fp16 = add(x = x_265_cast_fp16, y = linear_131_cast_fp16)[name = tensor<string, []>("x_271_cast_fp16")];
            tensor<int32, []> var_2476 = const()[name = tensor<string, []>("op_2476"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2493_axes_0 = const()[name = tensor<string, []>("op_2493_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_22_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880433728)))];
            tensor<fp16, [1280]> blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880436352)))];
            tensor<fp16, []> var_2482_to_fp16 = const()[name = tensor<string, []>("op_2482_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2493_cast_fp16 = layer_norm(axes = var_2493_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = tensor<string, []>("op_2493_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2504_to_fp16 = const()[name = tensor<string, []>("op_2504_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880438976)))];
            tensor<fp16, [1280]> var_2505_to_fp16 = const()[name = tensor<string, []>("op_2505_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883715840)))];
            tensor<fp16, [1, 1500, 1280]> linear_132_cast_fp16 = linear(bias = var_2505_to_fp16, weight = var_2504_to_fp16, x = var_2493_cast_fp16)[name = tensor<string, []>("linear_132_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2508_to_fp16 = const()[name = tensor<string, []>("op_2508_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883718464)))];
            tensor<fp16, [1, 1500, 1280]> linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2508_to_fp16, x = var_2493_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2512_to_fp16 = const()[name = tensor<string, []>("op_2512_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886995328)))];
            tensor<fp16, [1280]> var_2513_to_fp16 = const()[name = tensor<string, []>("op_2513_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(890272192)))];
            tensor<fp16, [1, 1500, 1280]> linear_134_cast_fp16 = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = var_2493_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")];
            tensor<int32, [4]> var_2521 = const()[name = tensor<string, []>("op_2521"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2522_cast_fp16 = reshape(shape = var_2521, x = linear_132_cast_fp16)[name = tensor<string, []>("op_2522_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_268_to_fp16 = const()[name = tensor<string, []>("const_268_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_91_cast_fp16 = mul(x = var_2522_cast_fp16, y = const_268_to_fp16)[name = tensor<string, []>("q_91_cast_fp16")];
            tensor<int32, [4]> var_2528 = const()[name = tensor<string, []>("op_2528"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2529_cast_fp16 = reshape(shape = var_2528, x = linear_133_cast_fp16)[name = tensor<string, []>("op_2529_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_269_to_fp16 = const()[name = tensor<string, []>("const_269_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_91_cast_fp16 = mul(x = var_2529_cast_fp16, y = const_269_to_fp16)[name = tensor<string, []>("k_91_cast_fp16")];
            tensor<int32, [4]> var_2535 = const()[name = tensor<string, []>("op_2535"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2536_cast_fp16 = reshape(shape = var_2535, x = linear_134_cast_fp16)[name = tensor<string, []>("op_2536_cast_fp16")];
            tensor<int32, [4]> var_2537 = const()[name = tensor<string, []>("op_2537"), 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_172_perm_0 = const()[name = tensor<string, []>("transpose_172_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_173_perm_0 = const()[name = tensor<string, []>("transpose_173_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_229 = transpose(perm = transpose_173_perm_0, x = k_91_cast_fp16)[name = tensor<string, []>("transpose_229")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_230 = transpose(perm = transpose_172_perm_0, x = q_91_cast_fp16)[name = tensor<string, []>("transpose_230")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_45_cast_fp16 = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_230, y = transpose_229)[name = tensor<string, []>("qk_45_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2541_cast_fp16 = softmax(axis = var_2476, x = qk_45_cast_fp16)[name = tensor<string, []>("op_2541_cast_fp16")];
            tensor<bool, []> var_2543_transpose_x_0 = const()[name = tensor<string, []>("op_2543_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2543_transpose_y_0 = const()[name = tensor<string, []>("op_2543_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_231 = transpose(perm = var_2537, x = var_2536_cast_fp16)[name = tensor<string, []>("transpose_231")];
            tensor<fp16, [1, 20, 1500, 64]> var_2543_cast_fp16 = matmul(transpose_x = var_2543_transpose_x_0, transpose_y = var_2543_transpose_y_0, x = var_2541_cast_fp16, y = transpose_231)[name = tensor<string, []>("op_2543_cast_fp16")];
            tensor<int32, [4]> var_2544 = const()[name = tensor<string, []>("op_2544"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_228 = transpose(perm = var_2544, x = var_2543_cast_fp16)[name = tensor<string, []>("transpose_228")];
            tensor<fp16, [1, 1500, 1280]> x_275_cast_fp16 = reshape(shape = concat_22, x = transpose_228)[name = tensor<string, []>("x_275_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2549_to_fp16 = const()[name = tensor<string, []>("op_2549_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(890274816)))];
            tensor<fp16, [1280]> var_2550_to_fp16 = const()[name = tensor<string, []>("op_2550_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893551680)))];
            tensor<fp16, [1, 1500, 1280]> linear_135_cast_fp16 = linear(bias = var_2550_to_fp16, weight = var_2549_to_fp16, x = x_275_cast_fp16)[name = tensor<string, []>("linear_135_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_277_cast_fp16 = add(x = x_271_cast_fp16, y = linear_135_cast_fp16)[name = tensor<string, []>("x_277_cast_fp16")];
            tensor<int32, [1]> var_2557_axes_0 = const()[name = tensor<string, []>("op_2557_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_22_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893554304)))];
            tensor<fp16, [1280]> blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893556928)))];
            tensor<fp16, [1, 1500, 1280]> var_2557_cast_fp16 = layer_norm(axes = var_2557_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = tensor<string, []>("op_2557_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2566_to_fp16 = const()[name = tensor<string, []>("op_2566_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893559552)))];
            tensor<fp16, [5120]> var_2567_to_fp16 = const()[name = tensor<string, []>("op_2567_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(906666816)))];
            tensor<fp16, [1, 1500, 5120]> linear_136_cast_fp16 = linear(bias = var_2567_to_fp16, weight = var_2566_to_fp16, x = var_2557_cast_fp16)[name = tensor<string, []>("linear_136_cast_fp16")];
            tensor<string, []> x_281_mode_0 = const()[name = tensor<string, []>("x_281_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_281_cast_fp16 = gelu(mode = x_281_mode_0, x = linear_136_cast_fp16)[name = tensor<string, []>("x_281_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2572_to_fp16 = const()[name = tensor<string, []>("op_2572_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(906677120)))];
            tensor<fp16, [1280]> var_2573_to_fp16 = const()[name = tensor<string, []>("op_2573_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919784384)))];
            tensor<fp16, [1, 1500, 1280]> linear_137_cast_fp16 = linear(bias = var_2573_to_fp16, weight = var_2572_to_fp16, x = x_281_cast_fp16)[name = tensor<string, []>("linear_137_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_283_cast_fp16 = add(x = x_277_cast_fp16, y = linear_137_cast_fp16)[name = tensor<string, []>("x_283_cast_fp16")];
            tensor<int32, []> var_2583 = const()[name = tensor<string, []>("op_2583"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2600_axes_0 = const()[name = tensor<string, []>("op_2600_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_23_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919787008)))];
            tensor<fp16, [1280]> blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919789632)))];
            tensor<fp16, []> var_2589_to_fp16 = const()[name = tensor<string, []>("op_2589_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2600_cast_fp16 = layer_norm(axes = var_2600_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = tensor<string, []>("op_2600_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2611_to_fp16 = const()[name = tensor<string, []>("op_2611_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919792256)))];
            tensor<fp16, [1280]> var_2612_to_fp16 = const()[name = tensor<string, []>("op_2612_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923069120)))];
            tensor<fp16, [1, 1500, 1280]> linear_138_cast_fp16 = linear(bias = var_2612_to_fp16, weight = var_2611_to_fp16, x = var_2600_cast_fp16)[name = tensor<string, []>("linear_138_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2615_to_fp16 = const()[name = tensor<string, []>("op_2615_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923071744)))];
            tensor<fp16, [1, 1500, 1280]> linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2615_to_fp16, x = var_2600_cast_fp16)[name = tensor<string, []>("linear_139_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2619_to_fp16 = const()[name = tensor<string, []>("op_2619_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926348608)))];
            tensor<fp16, [1280]> var_2620_to_fp16 = const()[name = tensor<string, []>("op_2620_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929625472)))];
            tensor<fp16, [1, 1500, 1280]> linear_140_cast_fp16 = linear(bias = var_2620_to_fp16, weight = var_2619_to_fp16, x = var_2600_cast_fp16)[name = tensor<string, []>("linear_140_cast_fp16")];
            tensor<int32, [4]> var_2628 = const()[name = tensor<string, []>("op_2628"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2629_cast_fp16 = reshape(shape = var_2628, x = linear_138_cast_fp16)[name = tensor<string, []>("op_2629_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_270_to_fp16 = const()[name = tensor<string, []>("const_270_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_95_cast_fp16 = mul(x = var_2629_cast_fp16, y = const_270_to_fp16)[name = tensor<string, []>("q_95_cast_fp16")];
            tensor<int32, [4]> var_2635 = const()[name = tensor<string, []>("op_2635"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2636_cast_fp16 = reshape(shape = var_2635, x = linear_139_cast_fp16)[name = tensor<string, []>("op_2636_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_271_to_fp16 = const()[name = tensor<string, []>("const_271_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_95_cast_fp16 = mul(x = var_2636_cast_fp16, y = const_271_to_fp16)[name = tensor<string, []>("k_95_cast_fp16")];
            tensor<int32, [4]> var_2642 = const()[name = tensor<string, []>("op_2642"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2643_cast_fp16 = reshape(shape = var_2642, x = linear_140_cast_fp16)[name = tensor<string, []>("op_2643_cast_fp16")];
            tensor<int32, [4]> var_2644 = const()[name = tensor<string, []>("op_2644"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_47_transpose_x_0 = const()[name = tensor<string, []>("qk_47_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_47_transpose_y_0 = const()[name = tensor<string, []>("qk_47_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_174_perm_0 = const()[name = tensor<string, []>("transpose_174_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_175_perm_0 = const()[name = tensor<string, []>("transpose_175_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_225 = transpose(perm = transpose_175_perm_0, x = k_95_cast_fp16)[name = tensor<string, []>("transpose_225")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_226 = transpose(perm = transpose_174_perm_0, x = q_95_cast_fp16)[name = tensor<string, []>("transpose_226")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_47_cast_fp16 = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = transpose_226, y = transpose_225)[name = tensor<string, []>("qk_47_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2648_cast_fp16 = softmax(axis = var_2583, x = qk_47_cast_fp16)[name = tensor<string, []>("op_2648_cast_fp16")];
            tensor<bool, []> var_2650_transpose_x_0 = const()[name = tensor<string, []>("op_2650_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2650_transpose_y_0 = const()[name = tensor<string, []>("op_2650_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_227 = transpose(perm = var_2644, x = var_2643_cast_fp16)[name = tensor<string, []>("transpose_227")];
            tensor<fp16, [1, 20, 1500, 64]> var_2650_cast_fp16 = matmul(transpose_x = var_2650_transpose_x_0, transpose_y = var_2650_transpose_y_0, x = var_2648_cast_fp16, y = transpose_227)[name = tensor<string, []>("op_2650_cast_fp16")];
            tensor<int32, [4]> var_2651 = const()[name = tensor<string, []>("op_2651"), 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, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_224 = transpose(perm = var_2651, x = var_2650_cast_fp16)[name = tensor<string, []>("transpose_224")];
            tensor<fp16, [1, 1500, 1280]> x_287_cast_fp16 = reshape(shape = concat_23, x = transpose_224)[name = tensor<string, []>("x_287_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2656_to_fp16 = const()[name = tensor<string, []>("op_2656_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929628096)))];
            tensor<fp16, [1280]> var_2657_to_fp16 = const()[name = tensor<string, []>("op_2657_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932904960)))];
            tensor<fp16, [1, 1500, 1280]> linear_141_cast_fp16 = linear(bias = var_2657_to_fp16, weight = var_2656_to_fp16, x = x_287_cast_fp16)[name = tensor<string, []>("linear_141_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_289_cast_fp16 = add(x = x_283_cast_fp16, y = linear_141_cast_fp16)[name = tensor<string, []>("x_289_cast_fp16")];
            tensor<int32, [1]> var_2664_axes_0 = const()[name = tensor<string, []>("op_2664_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_23_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932907584)))];
            tensor<fp16, [1280]> blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932910208)))];
            tensor<fp16, [1, 1500, 1280]> var_2664_cast_fp16 = layer_norm(axes = var_2664_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = tensor<string, []>("op_2664_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2673_to_fp16 = const()[name = tensor<string, []>("op_2673_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932912832)))];
            tensor<fp16, [5120]> var_2674_to_fp16 = const()[name = tensor<string, []>("op_2674_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(946020096)))];
            tensor<fp16, [1, 1500, 5120]> linear_142_cast_fp16 = linear(bias = var_2674_to_fp16, weight = var_2673_to_fp16, x = var_2664_cast_fp16)[name = tensor<string, []>("linear_142_cast_fp16")];
            tensor<string, []> x_293_mode_0 = const()[name = tensor<string, []>("x_293_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_293_cast_fp16 = gelu(mode = x_293_mode_0, x = linear_142_cast_fp16)[name = tensor<string, []>("x_293_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2679_to_fp16 = const()[name = tensor<string, []>("op_2679_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(946030400)))];
            tensor<fp16, [1280]> var_2680_to_fp16 = const()[name = tensor<string, []>("op_2680_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959137664)))];
            tensor<fp16, [1, 1500, 1280]> linear_143_cast_fp16 = linear(bias = var_2680_to_fp16, weight = var_2679_to_fp16, x = x_293_cast_fp16)[name = tensor<string, []>("linear_143_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_295_cast_fp16 = add(x = x_289_cast_fp16, y = linear_143_cast_fp16)[name = tensor<string, []>("x_295_cast_fp16")];
            tensor<int32, []> var_2690 = const()[name = tensor<string, []>("op_2690"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2707_axes_0 = const()[name = tensor<string, []>("op_2707_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_24_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_24_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959140288)))];
            tensor<fp16, [1280]> blocks_24_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_24_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959142912)))];
            tensor<fp16, []> var_2696_to_fp16 = const()[name = tensor<string, []>("op_2696_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2707_cast_fp16 = layer_norm(axes = var_2707_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast_fp16)[name = tensor<string, []>("op_2707_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2718_to_fp16 = const()[name = tensor<string, []>("op_2718_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959145536)))];
            tensor<fp16, [1280]> var_2719_to_fp16 = const()[name = tensor<string, []>("op_2719_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962422400)))];
            tensor<fp16, [1, 1500, 1280]> linear_144_cast_fp16 = linear(bias = var_2719_to_fp16, weight = var_2718_to_fp16, x = var_2707_cast_fp16)[name = tensor<string, []>("linear_144_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2722_to_fp16 = const()[name = tensor<string, []>("op_2722_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962425024)))];
            tensor<fp16, [1, 1500, 1280]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2722_to_fp16, x = var_2707_cast_fp16)[name = tensor<string, []>("linear_145_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2726_to_fp16 = const()[name = tensor<string, []>("op_2726_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(965701888)))];
            tensor<fp16, [1280]> var_2727_to_fp16 = const()[name = tensor<string, []>("op_2727_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(968978752)))];
            tensor<fp16, [1, 1500, 1280]> linear_146_cast_fp16 = linear(bias = var_2727_to_fp16, weight = var_2726_to_fp16, x = var_2707_cast_fp16)[name = tensor<string, []>("linear_146_cast_fp16")];
            tensor<int32, [4]> var_2735 = const()[name = tensor<string, []>("op_2735"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2736_cast_fp16 = reshape(shape = var_2735, x = linear_144_cast_fp16)[name = tensor<string, []>("op_2736_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_272_to_fp16 = const()[name = tensor<string, []>("const_272_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_99_cast_fp16 = mul(x = var_2736_cast_fp16, y = const_272_to_fp16)[name = tensor<string, []>("q_99_cast_fp16")];
            tensor<int32, [4]> var_2742 = const()[name = tensor<string, []>("op_2742"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2743_cast_fp16 = reshape(shape = var_2742, x = linear_145_cast_fp16)[name = tensor<string, []>("op_2743_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_273_to_fp16 = const()[name = tensor<string, []>("const_273_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_99_cast_fp16 = mul(x = var_2743_cast_fp16, y = const_273_to_fp16)[name = tensor<string, []>("k_99_cast_fp16")];
            tensor<int32, [4]> var_2749 = const()[name = tensor<string, []>("op_2749"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2750_cast_fp16 = reshape(shape = var_2749, x = linear_146_cast_fp16)[name = tensor<string, []>("op_2750_cast_fp16")];
            tensor<int32, [4]> var_2751 = const()[name = tensor<string, []>("op_2751"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_49_transpose_x_0 = const()[name = tensor<string, []>("qk_49_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_49_transpose_y_0 = const()[name = tensor<string, []>("qk_49_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_176_perm_0 = const()[name = tensor<string, []>("transpose_176_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_177_perm_0 = const()[name = tensor<string, []>("transpose_177_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_221 = transpose(perm = transpose_177_perm_0, x = k_99_cast_fp16)[name = tensor<string, []>("transpose_221")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_222 = transpose(perm = transpose_176_perm_0, x = q_99_cast_fp16)[name = tensor<string, []>("transpose_222")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_49_cast_fp16 = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = transpose_222, y = transpose_221)[name = tensor<string, []>("qk_49_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2755_cast_fp16 = softmax(axis = var_2690, x = qk_49_cast_fp16)[name = tensor<string, []>("op_2755_cast_fp16")];
            tensor<bool, []> var_2757_transpose_x_0 = const()[name = tensor<string, []>("op_2757_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2757_transpose_y_0 = const()[name = tensor<string, []>("op_2757_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_223 = transpose(perm = var_2751, x = var_2750_cast_fp16)[name = tensor<string, []>("transpose_223")];
            tensor<fp16, [1, 20, 1500, 64]> var_2757_cast_fp16 = matmul(transpose_x = var_2757_transpose_x_0, transpose_y = var_2757_transpose_y_0, x = var_2755_cast_fp16, y = transpose_223)[name = tensor<string, []>("op_2757_cast_fp16")];
            tensor<int32, [4]> var_2758 = const()[name = tensor<string, []>("op_2758"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_24 = const()[name = tensor<string, []>("concat_24"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_220 = transpose(perm = var_2758, x = var_2757_cast_fp16)[name = tensor<string, []>("transpose_220")];
            tensor<fp16, [1, 1500, 1280]> x_299_cast_fp16 = reshape(shape = concat_24, x = transpose_220)[name = tensor<string, []>("x_299_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2763_to_fp16 = const()[name = tensor<string, []>("op_2763_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(968981376)))];
            tensor<fp16, [1280]> var_2764_to_fp16 = const()[name = tensor<string, []>("op_2764_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(972258240)))];
            tensor<fp16, [1, 1500, 1280]> linear_147_cast_fp16 = linear(bias = var_2764_to_fp16, weight = var_2763_to_fp16, x = x_299_cast_fp16)[name = tensor<string, []>("linear_147_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_301_cast_fp16 = add(x = x_295_cast_fp16, y = linear_147_cast_fp16)[name = tensor<string, []>("x_301_cast_fp16")];
            tensor<int32, [1]> var_2771_axes_0 = const()[name = tensor<string, []>("op_2771_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_24_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_24_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(972260864)))];
            tensor<fp16, [1280]> blocks_24_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_24_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(972263488)))];
            tensor<fp16, [1, 1500, 1280]> var_2771_cast_fp16 = layer_norm(axes = var_2771_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast_fp16)[name = tensor<string, []>("op_2771_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2780_to_fp16 = const()[name = tensor<string, []>("op_2780_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(972266112)))];
            tensor<fp16, [5120]> var_2781_to_fp16 = const()[name = tensor<string, []>("op_2781_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(985373376)))];
            tensor<fp16, [1, 1500, 5120]> linear_148_cast_fp16 = linear(bias = var_2781_to_fp16, weight = var_2780_to_fp16, x = var_2771_cast_fp16)[name = tensor<string, []>("linear_148_cast_fp16")];
            tensor<string, []> x_305_mode_0 = const()[name = tensor<string, []>("x_305_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_305_cast_fp16 = gelu(mode = x_305_mode_0, x = linear_148_cast_fp16)[name = tensor<string, []>("x_305_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2786_to_fp16 = const()[name = tensor<string, []>("op_2786_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(985383680)))];
            tensor<fp16, [1280]> var_2787_to_fp16 = const()[name = tensor<string, []>("op_2787_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998490944)))];
            tensor<fp16, [1, 1500, 1280]> linear_149_cast_fp16 = linear(bias = var_2787_to_fp16, weight = var_2786_to_fp16, x = x_305_cast_fp16)[name = tensor<string, []>("linear_149_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_307_cast_fp16 = add(x = x_301_cast_fp16, y = linear_149_cast_fp16)[name = tensor<string, []>("x_307_cast_fp16")];
            tensor<int32, []> var_2797 = const()[name = tensor<string, []>("op_2797"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2814_axes_0 = const()[name = tensor<string, []>("op_2814_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_25_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_25_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998493568)))];
            tensor<fp16, [1280]> blocks_25_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_25_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998496192)))];
            tensor<fp16, []> var_2803_to_fp16 = const()[name = tensor<string, []>("op_2803_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2814_cast_fp16 = layer_norm(axes = var_2814_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast_fp16)[name = tensor<string, []>("op_2814_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2825_to_fp16 = const()[name = tensor<string, []>("op_2825_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998498816)))];
            tensor<fp16, [1280]> var_2826_to_fp16 = const()[name = tensor<string, []>("op_2826_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001775680)))];
            tensor<fp16, [1, 1500, 1280]> linear_150_cast_fp16 = linear(bias = var_2826_to_fp16, weight = var_2825_to_fp16, x = var_2814_cast_fp16)[name = tensor<string, []>("linear_150_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2829_to_fp16 = const()[name = tensor<string, []>("op_2829_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001778304)))];
            tensor<fp16, [1, 1500, 1280]> linear_151_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2829_to_fp16, x = var_2814_cast_fp16)[name = tensor<string, []>("linear_151_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2833_to_fp16 = const()[name = tensor<string, []>("op_2833_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1005055168)))];
            tensor<fp16, [1280]> var_2834_to_fp16 = const()[name = tensor<string, []>("op_2834_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1008332032)))];
            tensor<fp16, [1, 1500, 1280]> linear_152_cast_fp16 = linear(bias = var_2834_to_fp16, weight = var_2833_to_fp16, x = var_2814_cast_fp16)[name = tensor<string, []>("linear_152_cast_fp16")];
            tensor<int32, [4]> var_2842 = const()[name = tensor<string, []>("op_2842"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2843_cast_fp16 = reshape(shape = var_2842, x = linear_150_cast_fp16)[name = tensor<string, []>("op_2843_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_274_to_fp16 = const()[name = tensor<string, []>("const_274_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_103_cast_fp16 = mul(x = var_2843_cast_fp16, y = const_274_to_fp16)[name = tensor<string, []>("q_103_cast_fp16")];
            tensor<int32, [4]> var_2849 = const()[name = tensor<string, []>("op_2849"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2850_cast_fp16 = reshape(shape = var_2849, x = linear_151_cast_fp16)[name = tensor<string, []>("op_2850_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_275_to_fp16 = const()[name = tensor<string, []>("const_275_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_103_cast_fp16 = mul(x = var_2850_cast_fp16, y = const_275_to_fp16)[name = tensor<string, []>("k_103_cast_fp16")];
            tensor<int32, [4]> var_2856 = const()[name = tensor<string, []>("op_2856"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2857_cast_fp16 = reshape(shape = var_2856, x = linear_152_cast_fp16)[name = tensor<string, []>("op_2857_cast_fp16")];
            tensor<int32, [4]> var_2858 = const()[name = tensor<string, []>("op_2858"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_51_transpose_x_0 = const()[name = tensor<string, []>("qk_51_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_51_transpose_y_0 = const()[name = tensor<string, []>("qk_51_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_178_perm_0 = const()[name = tensor<string, []>("transpose_178_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_179_perm_0 = const()[name = tensor<string, []>("transpose_179_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_217 = transpose(perm = transpose_179_perm_0, x = k_103_cast_fp16)[name = tensor<string, []>("transpose_217")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_218 = transpose(perm = transpose_178_perm_0, x = q_103_cast_fp16)[name = tensor<string, []>("transpose_218")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_51_cast_fp16 = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = transpose_218, y = transpose_217)[name = tensor<string, []>("qk_51_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2862_cast_fp16 = softmax(axis = var_2797, x = qk_51_cast_fp16)[name = tensor<string, []>("op_2862_cast_fp16")];
            tensor<bool, []> var_2864_transpose_x_0 = const()[name = tensor<string, []>("op_2864_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2864_transpose_y_0 = const()[name = tensor<string, []>("op_2864_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_219 = transpose(perm = var_2858, x = var_2857_cast_fp16)[name = tensor<string, []>("transpose_219")];
            tensor<fp16, [1, 20, 1500, 64]> var_2864_cast_fp16 = matmul(transpose_x = var_2864_transpose_x_0, transpose_y = var_2864_transpose_y_0, x = var_2862_cast_fp16, y = transpose_219)[name = tensor<string, []>("op_2864_cast_fp16")];
            tensor<int32, [4]> var_2865 = const()[name = tensor<string, []>("op_2865"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_25 = const()[name = tensor<string, []>("concat_25"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_216 = transpose(perm = var_2865, x = var_2864_cast_fp16)[name = tensor<string, []>("transpose_216")];
            tensor<fp16, [1, 1500, 1280]> x_311_cast_fp16 = reshape(shape = concat_25, x = transpose_216)[name = tensor<string, []>("x_311_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2870_to_fp16 = const()[name = tensor<string, []>("op_2870_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1008334656)))];
            tensor<fp16, [1280]> var_2871_to_fp16 = const()[name = tensor<string, []>("op_2871_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011611520)))];
            tensor<fp16, [1, 1500, 1280]> linear_153_cast_fp16 = linear(bias = var_2871_to_fp16, weight = var_2870_to_fp16, x = x_311_cast_fp16)[name = tensor<string, []>("linear_153_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_313_cast_fp16 = add(x = x_307_cast_fp16, y = linear_153_cast_fp16)[name = tensor<string, []>("x_313_cast_fp16")];
            tensor<int32, [1]> var_2878_axes_0 = const()[name = tensor<string, []>("op_2878_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_25_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_25_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011614144)))];
            tensor<fp16, [1280]> blocks_25_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_25_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011616768)))];
            tensor<fp16, [1, 1500, 1280]> var_2878_cast_fp16 = layer_norm(axes = var_2878_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast_fp16)[name = tensor<string, []>("op_2878_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2887_to_fp16 = const()[name = tensor<string, []>("op_2887_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011619392)))];
            tensor<fp16, [5120]> var_2888_to_fp16 = const()[name = tensor<string, []>("op_2888_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024726656)))];
            tensor<fp16, [1, 1500, 5120]> linear_154_cast_fp16 = linear(bias = var_2888_to_fp16, weight = var_2887_to_fp16, x = var_2878_cast_fp16)[name = tensor<string, []>("linear_154_cast_fp16")];
            tensor<string, []> x_317_mode_0 = const()[name = tensor<string, []>("x_317_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_317_cast_fp16 = gelu(mode = x_317_mode_0, x = linear_154_cast_fp16)[name = tensor<string, []>("x_317_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_2893_to_fp16 = const()[name = tensor<string, []>("op_2893_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024736960)))];
            tensor<fp16, [1280]> var_2894_to_fp16 = const()[name = tensor<string, []>("op_2894_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037844224)))];
            tensor<fp16, [1, 1500, 1280]> linear_155_cast_fp16 = linear(bias = var_2894_to_fp16, weight = var_2893_to_fp16, x = x_317_cast_fp16)[name = tensor<string, []>("linear_155_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_319_cast_fp16 = add(x = x_313_cast_fp16, y = linear_155_cast_fp16)[name = tensor<string, []>("x_319_cast_fp16")];
            tensor<int32, []> var_2904 = const()[name = tensor<string, []>("op_2904"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2921_axes_0 = const()[name = tensor<string, []>("op_2921_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_26_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_26_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037846848)))];
            tensor<fp16, [1280]> blocks_26_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_26_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037849472)))];
            tensor<fp16, []> var_2910_to_fp16 = const()[name = tensor<string, []>("op_2910_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2921_cast_fp16 = layer_norm(axes = var_2921_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast_fp16)[name = tensor<string, []>("op_2921_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2932_to_fp16 = const()[name = tensor<string, []>("op_2932_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037852096)))];
            tensor<fp16, [1280]> var_2933_to_fp16 = const()[name = tensor<string, []>("op_2933_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041128960)))];
            tensor<fp16, [1, 1500, 1280]> linear_156_cast_fp16 = linear(bias = var_2933_to_fp16, weight = var_2932_to_fp16, x = var_2921_cast_fp16)[name = tensor<string, []>("linear_156_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2936_to_fp16 = const()[name = tensor<string, []>("op_2936_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041131584)))];
            tensor<fp16, [1, 1500, 1280]> linear_157_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2936_to_fp16, x = var_2921_cast_fp16)[name = tensor<string, []>("linear_157_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2940_to_fp16 = const()[name = tensor<string, []>("op_2940_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1044408448)))];
            tensor<fp16, [1280]> var_2941_to_fp16 = const()[name = tensor<string, []>("op_2941_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1047685312)))];
            tensor<fp16, [1, 1500, 1280]> linear_158_cast_fp16 = linear(bias = var_2941_to_fp16, weight = var_2940_to_fp16, x = var_2921_cast_fp16)[name = tensor<string, []>("linear_158_cast_fp16")];
            tensor<int32, [4]> var_2949 = const()[name = tensor<string, []>("op_2949"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2950_cast_fp16 = reshape(shape = var_2949, x = linear_156_cast_fp16)[name = tensor<string, []>("op_2950_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_276_to_fp16 = const()[name = tensor<string, []>("const_276_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_107_cast_fp16 = mul(x = var_2950_cast_fp16, y = const_276_to_fp16)[name = tensor<string, []>("q_107_cast_fp16")];
            tensor<int32, [4]> var_2956 = const()[name = tensor<string, []>("op_2956"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2957_cast_fp16 = reshape(shape = var_2956, x = linear_157_cast_fp16)[name = tensor<string, []>("op_2957_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_277_to_fp16 = const()[name = tensor<string, []>("const_277_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_107_cast_fp16 = mul(x = var_2957_cast_fp16, y = const_277_to_fp16)[name = tensor<string, []>("k_107_cast_fp16")];
            tensor<int32, [4]> var_2963 = const()[name = tensor<string, []>("op_2963"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2964_cast_fp16 = reshape(shape = var_2963, x = linear_158_cast_fp16)[name = tensor<string, []>("op_2964_cast_fp16")];
            tensor<int32, [4]> var_2965 = const()[name = tensor<string, []>("op_2965"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_53_transpose_x_0 = const()[name = tensor<string, []>("qk_53_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_53_transpose_y_0 = const()[name = tensor<string, []>("qk_53_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_180_perm_0 = const()[name = tensor<string, []>("transpose_180_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_181_perm_0 = const()[name = tensor<string, []>("transpose_181_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_213 = transpose(perm = transpose_181_perm_0, x = k_107_cast_fp16)[name = tensor<string, []>("transpose_213")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_214 = transpose(perm = transpose_180_perm_0, x = q_107_cast_fp16)[name = tensor<string, []>("transpose_214")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_53_cast_fp16 = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = transpose_214, y = transpose_213)[name = tensor<string, []>("qk_53_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_2969_cast_fp16 = softmax(axis = var_2904, x = qk_53_cast_fp16)[name = tensor<string, []>("op_2969_cast_fp16")];
            tensor<bool, []> var_2971_transpose_x_0 = const()[name = tensor<string, []>("op_2971_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2971_transpose_y_0 = const()[name = tensor<string, []>("op_2971_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_215 = transpose(perm = var_2965, x = var_2964_cast_fp16)[name = tensor<string, []>("transpose_215")];
            tensor<fp16, [1, 20, 1500, 64]> var_2971_cast_fp16 = matmul(transpose_x = var_2971_transpose_x_0, transpose_y = var_2971_transpose_y_0, x = var_2969_cast_fp16, y = transpose_215)[name = tensor<string, []>("op_2971_cast_fp16")];
            tensor<int32, [4]> var_2972 = const()[name = tensor<string, []>("op_2972"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_26 = const()[name = tensor<string, []>("concat_26"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_212 = transpose(perm = var_2972, x = var_2971_cast_fp16)[name = tensor<string, []>("transpose_212")];
            tensor<fp16, [1, 1500, 1280]> x_323_cast_fp16 = reshape(shape = concat_26, x = transpose_212)[name = tensor<string, []>("x_323_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2977_to_fp16 = const()[name = tensor<string, []>("op_2977_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1047687936)))];
            tensor<fp16, [1280]> var_2978_to_fp16 = const()[name = tensor<string, []>("op_2978_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050964800)))];
            tensor<fp16, [1, 1500, 1280]> linear_159_cast_fp16 = linear(bias = var_2978_to_fp16, weight = var_2977_to_fp16, x = x_323_cast_fp16)[name = tensor<string, []>("linear_159_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_325_cast_fp16 = add(x = x_319_cast_fp16, y = linear_159_cast_fp16)[name = tensor<string, []>("x_325_cast_fp16")];
            tensor<int32, [1]> var_2985_axes_0 = const()[name = tensor<string, []>("op_2985_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_26_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_26_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050967424)))];
            tensor<fp16, [1280]> blocks_26_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_26_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050970048)))];
            tensor<fp16, [1, 1500, 1280]> var_2985_cast_fp16 = layer_norm(axes = var_2985_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast_fp16)[name = tensor<string, []>("op_2985_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_2994_to_fp16 = const()[name = tensor<string, []>("op_2994_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050972672)))];
            tensor<fp16, [5120]> var_2995_to_fp16 = const()[name = tensor<string, []>("op_2995_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1064079936)))];
            tensor<fp16, [1, 1500, 5120]> linear_160_cast_fp16 = linear(bias = var_2995_to_fp16, weight = var_2994_to_fp16, x = var_2985_cast_fp16)[name = tensor<string, []>("linear_160_cast_fp16")];
            tensor<string, []> x_329_mode_0 = const()[name = tensor<string, []>("x_329_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_329_cast_fp16 = gelu(mode = x_329_mode_0, x = linear_160_cast_fp16)[name = tensor<string, []>("x_329_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_3000_to_fp16 = const()[name = tensor<string, []>("op_3000_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1064090240)))];
            tensor<fp16, [1280]> var_3001_to_fp16 = const()[name = tensor<string, []>("op_3001_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077197504)))];
            tensor<fp16, [1, 1500, 1280]> linear_161_cast_fp16 = linear(bias = var_3001_to_fp16, weight = var_3000_to_fp16, x = x_329_cast_fp16)[name = tensor<string, []>("linear_161_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_331_cast_fp16 = add(x = x_325_cast_fp16, y = linear_161_cast_fp16)[name = tensor<string, []>("x_331_cast_fp16")];
            tensor<int32, []> var_3011 = const()[name = tensor<string, []>("op_3011"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3028_axes_0 = const()[name = tensor<string, []>("op_3028_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_27_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_27_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077200128)))];
            tensor<fp16, [1280]> blocks_27_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_27_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077202752)))];
            tensor<fp16, []> var_3017_to_fp16 = const()[name = tensor<string, []>("op_3017_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3028_cast_fp16 = layer_norm(axes = var_3028_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast_fp16)[name = tensor<string, []>("op_3028_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3039_to_fp16 = const()[name = tensor<string, []>("op_3039_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077205376)))];
            tensor<fp16, [1280]> var_3040_to_fp16 = const()[name = tensor<string, []>("op_3040_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1080482240)))];
            tensor<fp16, [1, 1500, 1280]> linear_162_cast_fp16 = linear(bias = var_3040_to_fp16, weight = var_3039_to_fp16, x = var_3028_cast_fp16)[name = tensor<string, []>("linear_162_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3043_to_fp16 = const()[name = tensor<string, []>("op_3043_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1080484864)))];
            tensor<fp16, [1, 1500, 1280]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3043_to_fp16, x = var_3028_cast_fp16)[name = tensor<string, []>("linear_163_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3047_to_fp16 = const()[name = tensor<string, []>("op_3047_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1083761728)))];
            tensor<fp16, [1280]> var_3048_to_fp16 = const()[name = tensor<string, []>("op_3048_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1087038592)))];
            tensor<fp16, [1, 1500, 1280]> linear_164_cast_fp16 = linear(bias = var_3048_to_fp16, weight = var_3047_to_fp16, x = var_3028_cast_fp16)[name = tensor<string, []>("linear_164_cast_fp16")];
            tensor<int32, [4]> var_3056 = const()[name = tensor<string, []>("op_3056"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3057_cast_fp16 = reshape(shape = var_3056, x = linear_162_cast_fp16)[name = tensor<string, []>("op_3057_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_278_to_fp16 = const()[name = tensor<string, []>("const_278_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_111_cast_fp16 = mul(x = var_3057_cast_fp16, y = const_278_to_fp16)[name = tensor<string, []>("q_111_cast_fp16")];
            tensor<int32, [4]> var_3063 = const()[name = tensor<string, []>("op_3063"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3064_cast_fp16 = reshape(shape = var_3063, x = linear_163_cast_fp16)[name = tensor<string, []>("op_3064_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_279_to_fp16 = const()[name = tensor<string, []>("const_279_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_111_cast_fp16 = mul(x = var_3064_cast_fp16, y = const_279_to_fp16)[name = tensor<string, []>("k_111_cast_fp16")];
            tensor<int32, [4]> var_3070 = const()[name = tensor<string, []>("op_3070"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3071_cast_fp16 = reshape(shape = var_3070, x = linear_164_cast_fp16)[name = tensor<string, []>("op_3071_cast_fp16")];
            tensor<int32, [4]> var_3072 = const()[name = tensor<string, []>("op_3072"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_55_transpose_x_0 = const()[name = tensor<string, []>("qk_55_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_55_transpose_y_0 = const()[name = tensor<string, []>("qk_55_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_182_perm_0 = const()[name = tensor<string, []>("transpose_182_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_183_perm_0 = const()[name = tensor<string, []>("transpose_183_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_209 = transpose(perm = transpose_183_perm_0, x = k_111_cast_fp16)[name = tensor<string, []>("transpose_209")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_210 = transpose(perm = transpose_182_perm_0, x = q_111_cast_fp16)[name = tensor<string, []>("transpose_210")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_55_cast_fp16 = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = transpose_210, y = transpose_209)[name = tensor<string, []>("qk_55_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_3076_cast_fp16 = softmax(axis = var_3011, x = qk_55_cast_fp16)[name = tensor<string, []>("op_3076_cast_fp16")];
            tensor<bool, []> var_3078_transpose_x_0 = const()[name = tensor<string, []>("op_3078_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3078_transpose_y_0 = const()[name = tensor<string, []>("op_3078_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_211 = transpose(perm = var_3072, x = var_3071_cast_fp16)[name = tensor<string, []>("transpose_211")];
            tensor<fp16, [1, 20, 1500, 64]> var_3078_cast_fp16 = matmul(transpose_x = var_3078_transpose_x_0, transpose_y = var_3078_transpose_y_0, x = var_3076_cast_fp16, y = transpose_211)[name = tensor<string, []>("op_3078_cast_fp16")];
            tensor<int32, [4]> var_3079 = const()[name = tensor<string, []>("op_3079"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_27 = const()[name = tensor<string, []>("concat_27"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_208 = transpose(perm = var_3079, x = var_3078_cast_fp16)[name = tensor<string, []>("transpose_208")];
            tensor<fp16, [1, 1500, 1280]> x_335_cast_fp16 = reshape(shape = concat_27, x = transpose_208)[name = tensor<string, []>("x_335_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3084_to_fp16 = const()[name = tensor<string, []>("op_3084_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1087041216)))];
            tensor<fp16, [1280]> var_3085_to_fp16 = const()[name = tensor<string, []>("op_3085_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090318080)))];
            tensor<fp16, [1, 1500, 1280]> linear_165_cast_fp16 = linear(bias = var_3085_to_fp16, weight = var_3084_to_fp16, x = x_335_cast_fp16)[name = tensor<string, []>("linear_165_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_337_cast_fp16 = add(x = x_331_cast_fp16, y = linear_165_cast_fp16)[name = tensor<string, []>("x_337_cast_fp16")];
            tensor<int32, [1]> var_3092_axes_0 = const()[name = tensor<string, []>("op_3092_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_27_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_27_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090320704)))];
            tensor<fp16, [1280]> blocks_27_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_27_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090323328)))];
            tensor<fp16, [1, 1500, 1280]> var_3092_cast_fp16 = layer_norm(axes = var_3092_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast_fp16)[name = tensor<string, []>("op_3092_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_3101_to_fp16 = const()[name = tensor<string, []>("op_3101_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090325952)))];
            tensor<fp16, [5120]> var_3102_to_fp16 = const()[name = tensor<string, []>("op_3102_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1103433216)))];
            tensor<fp16, [1, 1500, 5120]> linear_166_cast_fp16 = linear(bias = var_3102_to_fp16, weight = var_3101_to_fp16, x = var_3092_cast_fp16)[name = tensor<string, []>("linear_166_cast_fp16")];
            tensor<string, []> x_341_mode_0 = const()[name = tensor<string, []>("x_341_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_341_cast_fp16 = gelu(mode = x_341_mode_0, x = linear_166_cast_fp16)[name = tensor<string, []>("x_341_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_3107_to_fp16 = const()[name = tensor<string, []>("op_3107_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1103443520)))];
            tensor<fp16, [1280]> var_3108_to_fp16 = const()[name = tensor<string, []>("op_3108_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116550784)))];
            tensor<fp16, [1, 1500, 1280]> linear_167_cast_fp16 = linear(bias = var_3108_to_fp16, weight = var_3107_to_fp16, x = x_341_cast_fp16)[name = tensor<string, []>("linear_167_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_343_cast_fp16 = add(x = x_337_cast_fp16, y = linear_167_cast_fp16)[name = tensor<string, []>("x_343_cast_fp16")];
            tensor<int32, []> var_3118 = const()[name = tensor<string, []>("op_3118"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3135_axes_0 = const()[name = tensor<string, []>("op_3135_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_28_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_28_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116553408)))];
            tensor<fp16, [1280]> blocks_28_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_28_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116556032)))];
            tensor<fp16, []> var_3124_to_fp16 = const()[name = tensor<string, []>("op_3124_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3135_cast_fp16 = layer_norm(axes = var_3135_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast_fp16)[name = tensor<string, []>("op_3135_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3146_to_fp16 = const()[name = tensor<string, []>("op_3146_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116558656)))];
            tensor<fp16, [1280]> var_3147_to_fp16 = const()[name = tensor<string, []>("op_3147_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1119835520)))];
            tensor<fp16, [1, 1500, 1280]> linear_168_cast_fp16 = linear(bias = var_3147_to_fp16, weight = var_3146_to_fp16, x = var_3135_cast_fp16)[name = tensor<string, []>("linear_168_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3150_to_fp16 = const()[name = tensor<string, []>("op_3150_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1119838144)))];
            tensor<fp16, [1, 1500, 1280]> linear_169_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3150_to_fp16, x = var_3135_cast_fp16)[name = tensor<string, []>("linear_169_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3154_to_fp16 = const()[name = tensor<string, []>("op_3154_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1123115008)))];
            tensor<fp16, [1280]> var_3155_to_fp16 = const()[name = tensor<string, []>("op_3155_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126391872)))];
            tensor<fp16, [1, 1500, 1280]> linear_170_cast_fp16 = linear(bias = var_3155_to_fp16, weight = var_3154_to_fp16, x = var_3135_cast_fp16)[name = tensor<string, []>("linear_170_cast_fp16")];
            tensor<int32, [4]> var_3163 = const()[name = tensor<string, []>("op_3163"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3164_cast_fp16 = reshape(shape = var_3163, x = linear_168_cast_fp16)[name = tensor<string, []>("op_3164_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_280_to_fp16 = const()[name = tensor<string, []>("const_280_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_115_cast_fp16 = mul(x = var_3164_cast_fp16, y = const_280_to_fp16)[name = tensor<string, []>("q_115_cast_fp16")];
            tensor<int32, [4]> var_3170 = const()[name = tensor<string, []>("op_3170"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3171_cast_fp16 = reshape(shape = var_3170, x = linear_169_cast_fp16)[name = tensor<string, []>("op_3171_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_281_to_fp16 = const()[name = tensor<string, []>("const_281_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_115_cast_fp16 = mul(x = var_3171_cast_fp16, y = const_281_to_fp16)[name = tensor<string, []>("k_115_cast_fp16")];
            tensor<int32, [4]> var_3177 = const()[name = tensor<string, []>("op_3177"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3178_cast_fp16 = reshape(shape = var_3177, x = linear_170_cast_fp16)[name = tensor<string, []>("op_3178_cast_fp16")];
            tensor<int32, [4]> var_3179 = const()[name = tensor<string, []>("op_3179"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_57_transpose_x_0 = const()[name = tensor<string, []>("qk_57_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_57_transpose_y_0 = const()[name = tensor<string, []>("qk_57_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_184_perm_0 = const()[name = tensor<string, []>("transpose_184_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_185_perm_0 = const()[name = tensor<string, []>("transpose_185_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_205 = transpose(perm = transpose_185_perm_0, x = k_115_cast_fp16)[name = tensor<string, []>("transpose_205")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_206 = transpose(perm = transpose_184_perm_0, x = q_115_cast_fp16)[name = tensor<string, []>("transpose_206")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_57_cast_fp16 = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = transpose_206, y = transpose_205)[name = tensor<string, []>("qk_57_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_3183_cast_fp16 = softmax(axis = var_3118, x = qk_57_cast_fp16)[name = tensor<string, []>("op_3183_cast_fp16")];
            tensor<bool, []> var_3185_transpose_x_0 = const()[name = tensor<string, []>("op_3185_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3185_transpose_y_0 = const()[name = tensor<string, []>("op_3185_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_207 = transpose(perm = var_3179, x = var_3178_cast_fp16)[name = tensor<string, []>("transpose_207")];
            tensor<fp16, [1, 20, 1500, 64]> var_3185_cast_fp16 = matmul(transpose_x = var_3185_transpose_x_0, transpose_y = var_3185_transpose_y_0, x = var_3183_cast_fp16, y = transpose_207)[name = tensor<string, []>("op_3185_cast_fp16")];
            tensor<int32, [4]> var_3186 = const()[name = tensor<string, []>("op_3186"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_28 = const()[name = tensor<string, []>("concat_28"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_204 = transpose(perm = var_3186, x = var_3185_cast_fp16)[name = tensor<string, []>("transpose_204")];
            tensor<fp16, [1, 1500, 1280]> x_347_cast_fp16 = reshape(shape = concat_28, x = transpose_204)[name = tensor<string, []>("x_347_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3191_to_fp16 = const()[name = tensor<string, []>("op_3191_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126394496)))];
            tensor<fp16, [1280]> var_3192_to_fp16 = const()[name = tensor<string, []>("op_3192_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129671360)))];
            tensor<fp16, [1, 1500, 1280]> linear_171_cast_fp16 = linear(bias = var_3192_to_fp16, weight = var_3191_to_fp16, x = x_347_cast_fp16)[name = tensor<string, []>("linear_171_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_349_cast_fp16 = add(x = x_343_cast_fp16, y = linear_171_cast_fp16)[name = tensor<string, []>("x_349_cast_fp16")];
            tensor<int32, [1]> var_3199_axes_0 = const()[name = tensor<string, []>("op_3199_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_28_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_28_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129673984)))];
            tensor<fp16, [1280]> blocks_28_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_28_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129676608)))];
            tensor<fp16, [1, 1500, 1280]> var_3199_cast_fp16 = layer_norm(axes = var_3199_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast_fp16)[name = tensor<string, []>("op_3199_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_3208_to_fp16 = const()[name = tensor<string, []>("op_3208_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129679232)))];
            tensor<fp16, [5120]> var_3209_to_fp16 = const()[name = tensor<string, []>("op_3209_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142786496)))];
            tensor<fp16, [1, 1500, 5120]> linear_172_cast_fp16 = linear(bias = var_3209_to_fp16, weight = var_3208_to_fp16, x = var_3199_cast_fp16)[name = tensor<string, []>("linear_172_cast_fp16")];
            tensor<string, []> x_353_mode_0 = const()[name = tensor<string, []>("x_353_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_353_cast_fp16 = gelu(mode = x_353_mode_0, x = linear_172_cast_fp16)[name = tensor<string, []>("x_353_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_3214_to_fp16 = const()[name = tensor<string, []>("op_3214_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142796800)))];
            tensor<fp16, [1280]> var_3215_to_fp16 = const()[name = tensor<string, []>("op_3215_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155904064)))];
            tensor<fp16, [1, 1500, 1280]> linear_173_cast_fp16 = linear(bias = var_3215_to_fp16, weight = var_3214_to_fp16, x = x_353_cast_fp16)[name = tensor<string, []>("linear_173_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_355_cast_fp16 = add(x = x_349_cast_fp16, y = linear_173_cast_fp16)[name = tensor<string, []>("x_355_cast_fp16")];
            tensor<int32, []> var_3225 = const()[name = tensor<string, []>("op_3225"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3242_axes_0 = const()[name = tensor<string, []>("op_3242_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_29_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_29_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155906688)))];
            tensor<fp16, [1280]> blocks_29_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_29_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155909312)))];
            tensor<fp16, []> var_3231_to_fp16 = const()[name = tensor<string, []>("op_3231_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3242_cast_fp16 = layer_norm(axes = var_3242_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast_fp16)[name = tensor<string, []>("op_3242_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3253_to_fp16 = const()[name = tensor<string, []>("op_3253_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155911936)))];
            tensor<fp16, [1280]> var_3254_to_fp16 = const()[name = tensor<string, []>("op_3254_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1159188800)))];
            tensor<fp16, [1, 1500, 1280]> linear_174_cast_fp16 = linear(bias = var_3254_to_fp16, weight = var_3253_to_fp16, x = var_3242_cast_fp16)[name = tensor<string, []>("linear_174_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3257_to_fp16 = const()[name = tensor<string, []>("op_3257_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1159191424)))];
            tensor<fp16, [1, 1500, 1280]> linear_175_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3257_to_fp16, x = var_3242_cast_fp16)[name = tensor<string, []>("linear_175_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3261_to_fp16 = const()[name = tensor<string, []>("op_3261_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162468288)))];
            tensor<fp16, [1280]> var_3262_to_fp16 = const()[name = tensor<string, []>("op_3262_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1165745152)))];
            tensor<fp16, [1, 1500, 1280]> linear_176_cast_fp16 = linear(bias = var_3262_to_fp16, weight = var_3261_to_fp16, x = var_3242_cast_fp16)[name = tensor<string, []>("linear_176_cast_fp16")];
            tensor<int32, [4]> var_3270 = const()[name = tensor<string, []>("op_3270"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3271_cast_fp16 = reshape(shape = var_3270, x = linear_174_cast_fp16)[name = tensor<string, []>("op_3271_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_282_to_fp16 = const()[name = tensor<string, []>("const_282_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_119_cast_fp16 = mul(x = var_3271_cast_fp16, y = const_282_to_fp16)[name = tensor<string, []>("q_119_cast_fp16")];
            tensor<int32, [4]> var_3277 = const()[name = tensor<string, []>("op_3277"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3278_cast_fp16 = reshape(shape = var_3277, x = linear_175_cast_fp16)[name = tensor<string, []>("op_3278_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_283_to_fp16 = const()[name = tensor<string, []>("const_283_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_119_cast_fp16 = mul(x = var_3278_cast_fp16, y = const_283_to_fp16)[name = tensor<string, []>("k_119_cast_fp16")];
            tensor<int32, [4]> var_3284 = const()[name = tensor<string, []>("op_3284"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3285_cast_fp16 = reshape(shape = var_3284, x = linear_176_cast_fp16)[name = tensor<string, []>("op_3285_cast_fp16")];
            tensor<int32, [4]> var_3286 = const()[name = tensor<string, []>("op_3286"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_59_transpose_x_0 = const()[name = tensor<string, []>("qk_59_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_59_transpose_y_0 = const()[name = tensor<string, []>("qk_59_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_186_perm_0 = const()[name = tensor<string, []>("transpose_186_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_187_perm_0 = const()[name = tensor<string, []>("transpose_187_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_201 = transpose(perm = transpose_187_perm_0, x = k_119_cast_fp16)[name = tensor<string, []>("transpose_201")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_202 = transpose(perm = transpose_186_perm_0, x = q_119_cast_fp16)[name = tensor<string, []>("transpose_202")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_59_cast_fp16 = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = transpose_202, y = transpose_201)[name = tensor<string, []>("qk_59_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_3290_cast_fp16 = softmax(axis = var_3225, x = qk_59_cast_fp16)[name = tensor<string, []>("op_3290_cast_fp16")];
            tensor<bool, []> var_3292_transpose_x_0 = const()[name = tensor<string, []>("op_3292_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3292_transpose_y_0 = const()[name = tensor<string, []>("op_3292_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_203 = transpose(perm = var_3286, x = var_3285_cast_fp16)[name = tensor<string, []>("transpose_203")];
            tensor<fp16, [1, 20, 1500, 64]> var_3292_cast_fp16 = matmul(transpose_x = var_3292_transpose_x_0, transpose_y = var_3292_transpose_y_0, x = var_3290_cast_fp16, y = transpose_203)[name = tensor<string, []>("op_3292_cast_fp16")];
            tensor<int32, [4]> var_3293 = const()[name = tensor<string, []>("op_3293"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_29 = const()[name = tensor<string, []>("concat_29"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_200 = transpose(perm = var_3293, x = var_3292_cast_fp16)[name = tensor<string, []>("transpose_200")];
            tensor<fp16, [1, 1500, 1280]> x_359_cast_fp16 = reshape(shape = concat_29, x = transpose_200)[name = tensor<string, []>("x_359_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3298_to_fp16 = const()[name = tensor<string, []>("op_3298_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1165747776)))];
            tensor<fp16, [1280]> var_3299_to_fp16 = const()[name = tensor<string, []>("op_3299_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1169024640)))];
            tensor<fp16, [1, 1500, 1280]> linear_177_cast_fp16 = linear(bias = var_3299_to_fp16, weight = var_3298_to_fp16, x = x_359_cast_fp16)[name = tensor<string, []>("linear_177_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_361_cast_fp16 = add(x = x_355_cast_fp16, y = linear_177_cast_fp16)[name = tensor<string, []>("x_361_cast_fp16")];
            tensor<int32, [1]> var_3306_axes_0 = const()[name = tensor<string, []>("op_3306_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_29_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_29_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1169027264)))];
            tensor<fp16, [1280]> blocks_29_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_29_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1169029888)))];
            tensor<fp16, [1, 1500, 1280]> var_3306_cast_fp16 = layer_norm(axes = var_3306_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast_fp16)[name = tensor<string, []>("op_3306_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_3315_to_fp16 = const()[name = tensor<string, []>("op_3315_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1169032512)))];
            tensor<fp16, [5120]> var_3316_to_fp16 = const()[name = tensor<string, []>("op_3316_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1182139776)))];
            tensor<fp16, [1, 1500, 5120]> linear_178_cast_fp16 = linear(bias = var_3316_to_fp16, weight = var_3315_to_fp16, x = var_3306_cast_fp16)[name = tensor<string, []>("linear_178_cast_fp16")];
            tensor<string, []> x_365_mode_0 = const()[name = tensor<string, []>("x_365_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_365_cast_fp16 = gelu(mode = x_365_mode_0, x = linear_178_cast_fp16)[name = tensor<string, []>("x_365_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_3321_to_fp16 = const()[name = tensor<string, []>("op_3321_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1182150080)))];
            tensor<fp16, [1280]> var_3322_to_fp16 = const()[name = tensor<string, []>("op_3322_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195257344)))];
            tensor<fp16, [1, 1500, 1280]> linear_179_cast_fp16 = linear(bias = var_3322_to_fp16, weight = var_3321_to_fp16, x = x_365_cast_fp16)[name = tensor<string, []>("linear_179_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_367_cast_fp16 = add(x = x_361_cast_fp16, y = linear_179_cast_fp16)[name = tensor<string, []>("x_367_cast_fp16")];
            tensor<int32, []> var_3332 = const()[name = tensor<string, []>("op_3332"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3349_axes_0 = const()[name = tensor<string, []>("op_3349_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_30_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_30_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195259968)))];
            tensor<fp16, [1280]> blocks_30_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_30_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195262592)))];
            tensor<fp16, []> var_3338_to_fp16 = const()[name = tensor<string, []>("op_3338_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3349_cast_fp16 = layer_norm(axes = var_3349_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast_fp16)[name = tensor<string, []>("op_3349_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3360_to_fp16 = const()[name = tensor<string, []>("op_3360_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195265216)))];
            tensor<fp16, [1280]> var_3361_to_fp16 = const()[name = tensor<string, []>("op_3361_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1198542080)))];
            tensor<fp16, [1, 1500, 1280]> linear_180_cast_fp16 = linear(bias = var_3361_to_fp16, weight = var_3360_to_fp16, x = var_3349_cast_fp16)[name = tensor<string, []>("linear_180_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3364_to_fp16 = const()[name = tensor<string, []>("op_3364_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1198544704)))];
            tensor<fp16, [1, 1500, 1280]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3364_to_fp16, x = var_3349_cast_fp16)[name = tensor<string, []>("linear_181_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3368_to_fp16 = const()[name = tensor<string, []>("op_3368_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1201821568)))];
            tensor<fp16, [1280]> var_3369_to_fp16 = const()[name = tensor<string, []>("op_3369_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1205098432)))];
            tensor<fp16, [1, 1500, 1280]> linear_182_cast_fp16 = linear(bias = var_3369_to_fp16, weight = var_3368_to_fp16, x = var_3349_cast_fp16)[name = tensor<string, []>("linear_182_cast_fp16")];
            tensor<int32, [4]> var_3377 = const()[name = tensor<string, []>("op_3377"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3378_cast_fp16 = reshape(shape = var_3377, x = linear_180_cast_fp16)[name = tensor<string, []>("op_3378_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_284_to_fp16 = const()[name = tensor<string, []>("const_284_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_123_cast_fp16 = mul(x = var_3378_cast_fp16, y = const_284_to_fp16)[name = tensor<string, []>("q_123_cast_fp16")];
            tensor<int32, [4]> var_3384 = const()[name = tensor<string, []>("op_3384"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3385_cast_fp16 = reshape(shape = var_3384, x = linear_181_cast_fp16)[name = tensor<string, []>("op_3385_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_285_to_fp16 = const()[name = tensor<string, []>("const_285_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_123_cast_fp16 = mul(x = var_3385_cast_fp16, y = const_285_to_fp16)[name = tensor<string, []>("k_123_cast_fp16")];
            tensor<int32, [4]> var_3391 = const()[name = tensor<string, []>("op_3391"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3392_cast_fp16 = reshape(shape = var_3391, x = linear_182_cast_fp16)[name = tensor<string, []>("op_3392_cast_fp16")];
            tensor<int32, [4]> var_3393 = const()[name = tensor<string, []>("op_3393"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_61_transpose_x_0 = const()[name = tensor<string, []>("qk_61_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_61_transpose_y_0 = const()[name = tensor<string, []>("qk_61_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_188_perm_0 = const()[name = tensor<string, []>("transpose_188_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_189_perm_0 = const()[name = tensor<string, []>("transpose_189_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_197 = transpose(perm = transpose_189_perm_0, x = k_123_cast_fp16)[name = tensor<string, []>("transpose_197")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_198 = transpose(perm = transpose_188_perm_0, x = q_123_cast_fp16)[name = tensor<string, []>("transpose_198")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_61_cast_fp16 = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = transpose_198, y = transpose_197)[name = tensor<string, []>("qk_61_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_3397_cast_fp16 = softmax(axis = var_3332, x = qk_61_cast_fp16)[name = tensor<string, []>("op_3397_cast_fp16")];
            tensor<bool, []> var_3399_transpose_x_0 = const()[name = tensor<string, []>("op_3399_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3399_transpose_y_0 = const()[name = tensor<string, []>("op_3399_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_199 = transpose(perm = var_3393, x = var_3392_cast_fp16)[name = tensor<string, []>("transpose_199")];
            tensor<fp16, [1, 20, 1500, 64]> var_3399_cast_fp16 = matmul(transpose_x = var_3399_transpose_x_0, transpose_y = var_3399_transpose_y_0, x = var_3397_cast_fp16, y = transpose_199)[name = tensor<string, []>("op_3399_cast_fp16")];
            tensor<int32, [4]> var_3400 = const()[name = tensor<string, []>("op_3400"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_30 = const()[name = tensor<string, []>("concat_30"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_196 = transpose(perm = var_3400, x = var_3399_cast_fp16)[name = tensor<string, []>("transpose_196")];
            tensor<fp16, [1, 1500, 1280]> x_371_cast_fp16 = reshape(shape = concat_30, x = transpose_196)[name = tensor<string, []>("x_371_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3405_to_fp16 = const()[name = tensor<string, []>("op_3405_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1205101056)))];
            tensor<fp16, [1280]> var_3406_to_fp16 = const()[name = tensor<string, []>("op_3406_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208377920)))];
            tensor<fp16, [1, 1500, 1280]> linear_183_cast_fp16 = linear(bias = var_3406_to_fp16, weight = var_3405_to_fp16, x = x_371_cast_fp16)[name = tensor<string, []>("linear_183_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_373_cast_fp16 = add(x = x_367_cast_fp16, y = linear_183_cast_fp16)[name = tensor<string, []>("x_373_cast_fp16")];
            tensor<int32, [1]> var_3413_axes_0 = const()[name = tensor<string, []>("op_3413_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_30_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_30_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208380544)))];
            tensor<fp16, [1280]> blocks_30_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_30_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208383168)))];
            tensor<fp16, [1, 1500, 1280]> var_3413_cast_fp16 = layer_norm(axes = var_3413_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast_fp16)[name = tensor<string, []>("op_3413_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_3422_to_fp16 = const()[name = tensor<string, []>("op_3422_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208385792)))];
            tensor<fp16, [5120]> var_3423_to_fp16 = const()[name = tensor<string, []>("op_3423_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221493056)))];
            tensor<fp16, [1, 1500, 5120]> linear_184_cast_fp16 = linear(bias = var_3423_to_fp16, weight = var_3422_to_fp16, x = var_3413_cast_fp16)[name = tensor<string, []>("linear_184_cast_fp16")];
            tensor<string, []> x_377_mode_0 = const()[name = tensor<string, []>("x_377_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_377_cast_fp16 = gelu(mode = x_377_mode_0, x = linear_184_cast_fp16)[name = tensor<string, []>("x_377_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_3428_to_fp16 = const()[name = tensor<string, []>("op_3428_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221503360)))];
            tensor<fp16, [1280]> var_3429_to_fp16 = const()[name = tensor<string, []>("op_3429_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234610624)))];
            tensor<fp16, [1, 1500, 1280]> linear_185_cast_fp16 = linear(bias = var_3429_to_fp16, weight = var_3428_to_fp16, x = x_377_cast_fp16)[name = tensor<string, []>("linear_185_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_379_cast_fp16 = add(x = x_373_cast_fp16, y = linear_185_cast_fp16)[name = tensor<string, []>("x_379_cast_fp16")];
            tensor<int32, []> var_3439 = const()[name = tensor<string, []>("op_3439"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3456_axes_0 = const()[name = tensor<string, []>("op_3456_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_31_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_31_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234613248)))];
            tensor<fp16, [1280]> blocks_31_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_31_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234615872)))];
            tensor<fp16, []> var_3445_to_fp16 = const()[name = tensor<string, []>("op_3445_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3456_cast_fp16 = layer_norm(axes = var_3456_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast_fp16)[name = tensor<string, []>("op_3456_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3467_to_fp16 = const()[name = tensor<string, []>("op_3467_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234618496)))];
            tensor<fp16, [1280]> var_3468_to_fp16 = const()[name = tensor<string, []>("op_3468_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1237895360)))];
            tensor<fp16, [1, 1500, 1280]> linear_186_cast_fp16 = linear(bias = var_3468_to_fp16, weight = var_3467_to_fp16, x = var_3456_cast_fp16)[name = tensor<string, []>("linear_186_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3471_to_fp16 = const()[name = tensor<string, []>("op_3471_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1237897984)))];
            tensor<fp16, [1, 1500, 1280]> linear_187_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3471_to_fp16, x = var_3456_cast_fp16)[name = tensor<string, []>("linear_187_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3475_to_fp16 = const()[name = tensor<string, []>("op_3475_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1241174848)))];
            tensor<fp16, [1280]> var_3476_to_fp16 = const()[name = tensor<string, []>("op_3476_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1244451712)))];
            tensor<fp16, [1, 1500, 1280]> linear_188_cast_fp16 = linear(bias = var_3476_to_fp16, weight = var_3475_to_fp16, x = var_3456_cast_fp16)[name = tensor<string, []>("linear_188_cast_fp16")];
            tensor<int32, [4]> var_3484 = const()[name = tensor<string, []>("op_3484"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3485_cast_fp16 = reshape(shape = var_3484, x = linear_186_cast_fp16)[name = tensor<string, []>("op_3485_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_286_to_fp16 = const()[name = tensor<string, []>("const_286_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> q_cast_fp16 = mul(x = var_3485_cast_fp16, y = const_286_to_fp16)[name = tensor<string, []>("q_cast_fp16")];
            tensor<int32, [4]> var_3491 = const()[name = tensor<string, []>("op_3491"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3492_cast_fp16 = reshape(shape = var_3491, x = linear_187_cast_fp16)[name = tensor<string, []>("op_3492_cast_fp16")];
            tensor<fp16, [1, 1, 1, 1]> const_287_to_fp16 = const()[name = tensor<string, []>("const_287_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])];
            tensor<fp16, [1, 1500, 20, 64]> k_cast_fp16 = mul(x = var_3492_cast_fp16, y = const_287_to_fp16)[name = tensor<string, []>("k_cast_fp16")];
            tensor<int32, [4]> var_3498 = const()[name = tensor<string, []>("op_3498"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3499_cast_fp16 = reshape(shape = var_3498, x = linear_188_cast_fp16)[name = tensor<string, []>("op_3499_cast_fp16")];
            tensor<int32, [4]> var_3500 = const()[name = tensor<string, []>("op_3500"), 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_190_perm_0 = const()[name = tensor<string, []>("transpose_190_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> transpose_191_perm_0 = const()[name = tensor<string, []>("transpose_191_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<fp16, [1, 20, 64, 1500]> transpose_193 = transpose(perm = transpose_191_perm_0, x = k_cast_fp16)[name = tensor<string, []>("transpose_193")];
            tensor<fp16, [1, 20, 1500, 64]> transpose_194 = transpose(perm = transpose_190_perm_0, x = q_cast_fp16)[name = tensor<string, []>("transpose_194")];
            tensor<fp16, [1, 20, 1500, 1500]> qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_194, y = transpose_193)[name = tensor<string, []>("qk_cast_fp16")];
            tensor<fp16, [1, 20, 1500, 1500]> var_3504_cast_fp16 = softmax(axis = var_3439, x = qk_cast_fp16)[name = tensor<string, []>("op_3504_cast_fp16")];
            tensor<bool, []> var_3506_transpose_x_0 = const()[name = tensor<string, []>("op_3506_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3506_transpose_y_0 = const()[name = tensor<string, []>("op_3506_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_195 = transpose(perm = var_3500, x = var_3499_cast_fp16)[name = tensor<string, []>("transpose_195")];
            tensor<fp16, [1, 20, 1500, 64]> var_3506_cast_fp16 = matmul(transpose_x = var_3506_transpose_x_0, transpose_y = var_3506_transpose_y_0, x = var_3504_cast_fp16, y = transpose_195)[name = tensor<string, []>("op_3506_cast_fp16")];
            tensor<int32, [4]> var_3507 = const()[name = tensor<string, []>("op_3507"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_31 = const()[name = tensor<string, []>("concat_31"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_192 = transpose(perm = var_3507, x = var_3506_cast_fp16)[name = tensor<string, []>("transpose_192")];
            tensor<fp16, [1, 1500, 1280]> x_383_cast_fp16 = reshape(shape = concat_31, x = transpose_192)[name = tensor<string, []>("x_383_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3512_to_fp16 = const()[name = tensor<string, []>("op_3512_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1244454336)))];
            tensor<fp16, [1280]> var_3513_to_fp16 = const()[name = tensor<string, []>("op_3513_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247731200)))];
            tensor<fp16, [1, 1500, 1280]> linear_189_cast_fp16 = linear(bias = var_3513_to_fp16, weight = var_3512_to_fp16, x = x_383_cast_fp16)[name = tensor<string, []>("linear_189_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_385_cast_fp16 = add(x = x_379_cast_fp16, y = linear_189_cast_fp16)[name = tensor<string, []>("x_385_cast_fp16")];
            tensor<int32, [1]> var_3520_axes_0 = const()[name = tensor<string, []>("op_3520_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> blocks_31_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_31_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247733824)))];
            tensor<fp16, [1280]> blocks_31_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_31_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247736448)))];
            tensor<fp16, [1, 1500, 1280]> var_3520_cast_fp16 = layer_norm(axes = var_3520_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast_fp16)[name = tensor<string, []>("op_3520_cast_fp16")];
            tensor<fp16, [5120, 1280]> var_3529_to_fp16 = const()[name = tensor<string, []>("op_3529_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247739072)))];
            tensor<fp16, [5120]> var_3530_to_fp16 = const()[name = tensor<string, []>("op_3530_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1260846336)))];
            tensor<fp16, [1, 1500, 5120]> linear_190_cast_fp16 = linear(bias = var_3530_to_fp16, weight = var_3529_to_fp16, x = var_3520_cast_fp16)[name = tensor<string, []>("linear_190_cast_fp16")];
            tensor<string, []> x_389_mode_0 = const()[name = tensor<string, []>("x_389_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 5120]> x_389_cast_fp16 = gelu(mode = x_389_mode_0, x = linear_190_cast_fp16)[name = tensor<string, []>("x_389_cast_fp16")];
            tensor<fp16, [1280, 5120]> var_3535_to_fp16 = const()[name = tensor<string, []>("op_3535_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1260856640)))];
            tensor<fp16, [1280]> var_3536_to_fp16 = const()[name = tensor<string, []>("op_3536_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273963904)))];
            tensor<fp16, [1, 1500, 1280]> linear_191_cast_fp16 = linear(bias = var_3536_to_fp16, weight = var_3535_to_fp16, x = x_389_cast_fp16)[name = tensor<string, []>("linear_191_cast_fp16")];
            tensor<fp16, [1, 1500, 1280]> x_cast_fp16 = add(x = x_385_cast_fp16, y = linear_191_cast_fp16)[name = tensor<string, []>("x_cast_fp16")];
            tensor<int32, [1]> var_3550_axes_0 = const()[name = tensor<string, []>("op_3550_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1280]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273966528)))];
            tensor<fp16, [1280]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273969152)))];
            tensor<fp16, []> var_3541_to_fp16 = const()[name = tensor<string, []>("op_3541_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3550_cast_fp16 = layer_norm(axes = var_3550_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3541_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = tensor<string, []>("op_3550_cast_fp16")];
            tensor<string, []> var_3550_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_3550_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [1, 1500, 1280]> output = cast(dtype = var_3550_cast_fp16_to_fp32_dtype_0, x = var_3550_cast_fp16)[name = tensor<string, []>("cast_192")];
        } -> (output);
}