program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}, {"coremltools-component-torch", "2.3.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.2"}})]
{
    func main<ios17>(tensor<fp16, [1, 80, 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<fp16, [1280, 80, 3]> weight_3_to_fp16 = const()[name = tensor<string, []>("weight_3_to_fp16"), val = tensor<fp16, [1280, 80, 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, []>(614528)))];
            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 = logmel_data)[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_89 = const()[name = tensor<string, []>("op_89"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> var_98 = const()[name = tensor<string, []>("op_98"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1]> var_100 = const()[name = tensor<string, []>("op_100"), val = tensor<int32, [1]>([1])];
            tensor<string, []> var_102_pad_type_0 = const()[name = tensor<string, []>("op_102_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> var_102_pad_0 = const()[name = tensor<string, []>("op_102_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, []>(617152)))];
            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, []>(10447616)))];
            tensor<fp16, [1, 1280, 1500]> var_102_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_100, groups = var_89, pad = var_102_pad_0, pad_type = var_102_pad_type_0, strides = var_98, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("op_102_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_102_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")];
            tensor<int32, [3]> var_108 = const()[name = tensor<string, []>("op_108"), 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, []>(10450240)))];
            tensor<fp16, [1, 1500, 1280]> transpose_320 = transpose(perm = var_108, x = x_3_cast_fp16)[name = tensor<string, []>("transpose_320")];
            tensor<fp16, [1, 1500, 1280]> var_111_cast_fp16 = add(x = transpose_320, y = positional_embedding_to_fp16)[name = tensor<string, []>("op_111_cast_fp16")];
            tensor<int32, []> var_124 = const()[name = tensor<string, []>("op_124"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_140_axes_0 = const()[name = tensor<string, []>("op_140_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, []>(14290304)))];
            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, []>(14292928)))];
            tensor<fp16, []> var_130_to_fp16 = const()[name = tensor<string, []>("op_130_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_140_cast_fp16 = layer_norm(axes = var_140_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_130_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_111_cast_fp16)[name = tensor<string, []>("op_140_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_151_to_fp16 = const()[name = tensor<string, []>("op_151_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14295552)))];
            tensor<fp16, [1280]> var_152_to_fp16 = const()[name = tensor<string, []>("op_152_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17572416)))];
            tensor<fp16, [1, 1500, 1280]> linear_0_cast_fp16 = linear(bias = var_152_to_fp16, weight = var_151_to_fp16, x = var_140_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_155_to_fp16 = const()[name = tensor<string, []>("op_155_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17575040)))];
            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, []>(20851904)))];
            tensor<fp16, [1, 1500, 1280]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_155_to_fp16, x = var_140_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_159_to_fp16 = const()[name = tensor<string, []>("op_159_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20854528)))];
            tensor<fp16, [1280]> var_160_to_fp16 = const()[name = tensor<string, []>("op_160_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24131392)))];
            tensor<fp16, [1, 1500, 1280]> linear_2_cast_fp16 = linear(bias = var_160_to_fp16, weight = var_159_to_fp16, x = var_140_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> var_168 = const()[name = tensor<string, []>("op_168"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_169_cast_fp16 = reshape(shape = var_168, x = linear_0_cast_fp16)[name = tensor<string, []>("op_169_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_169_cast_fp16, y = const_224_to_fp16)[name = tensor<string, []>("q_3_cast_fp16")];
            tensor<int32, [4]> var_175 = const()[name = tensor<string, []>("op_175"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_176_cast_fp16 = reshape(shape = var_175, x = linear_1_cast_fp16)[name = tensor<string, []>("op_176_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_176_cast_fp16, y = const_225_to_fp16)[name = tensor<string, []>("k_3_cast_fp16")];
            tensor<int32, [4]> var_182 = const()[name = tensor<string, []>("op_182"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_183_cast_fp16 = reshape(shape = var_182, x = linear_2_cast_fp16)[name = tensor<string, []>("op_183_cast_fp16")];
            tensor<int32, [4]> var_184 = const()[name = tensor<string, []>("op_184"), 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, -3, -1])];
            tensor<int32, [4]> transpose_129_perm_0 = const()[name = tensor<string, []>("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_188_cast_fp16 = softmax(axis = var_124, x = qk_1_cast_fp16)[name = tensor<string, []>("op_188_cast_fp16")];
            tensor<bool, []> var_190_transpose_x_0 = const()[name = tensor<string, []>("op_190_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_190_transpose_y_0 = const()[name = tensor<string, []>("op_190_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_319 = transpose(perm = var_184, x = var_183_cast_fp16)[name = tensor<string, []>("transpose_319")];
            tensor<fp16, [1, 20, 1500, 64]> var_190_cast_fp16 = matmul(transpose_x = var_190_transpose_x_0, transpose_y = var_190_transpose_y_0, x = var_188_cast_fp16, y = transpose_319)[name = tensor<string, []>("op_190_cast_fp16")];
            tensor<int32, [4]> var_191 = const()[name = tensor<string, []>("op_191"), 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_191, x = var_190_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_196_to_fp16 = const()[name = tensor<string, []>("op_196_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24134016)))];
            tensor<fp16, [1280]> var_197_to_fp16 = const()[name = tensor<string, []>("op_197_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27410880)))];
            tensor<fp16, [1, 1500, 1280]> linear_3_cast_fp16 = linear(bias = var_197_to_fp16, weight = var_196_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_111_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")];
            tensor<int32, [1]> var_204_axes_0 = const()[name = tensor<string, []>("op_204_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, []>(27413504)))];
            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, []>(27416128)))];
            tensor<fp16, [1, 1500, 1280]> var_204_cast_fp16 = layer_norm(axes = var_204_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_130_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor<string, []>("op_204_cast_fp16")];
            tensor<fp16, [2624, 1280]> var_213_to_fp16 = const()[name = tensor<string, []>("op_213_to_fp16"), val = tensor<fp16, [2624, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27418752)))];
            tensor<fp16, [2624]> var_214_to_fp16 = const()[name = tensor<string, []>("op_214_to_fp16"), val = tensor<fp16, [2624]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34136256)))];
            tensor<fp16, [1, 1500, 2624]> linear_4_cast_fp16 = linear(bias = var_214_to_fp16, weight = var_213_to_fp16, x = var_204_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, 2624]> 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, 2624]> var_219_to_fp16 = const()[name = tensor<string, []>("op_219_to_fp16"), val = tensor<fp16, [1280, 2624]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34141568)))];
            tensor<fp16, [1280]> var_220_to_fp16 = const()[name = tensor<string, []>("op_220_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40859072)))];
            tensor<fp16, [1, 1500, 1280]> linear_5_cast_fp16 = linear(bias = var_220_to_fp16, weight = var_219_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_230 = const()[name = tensor<string, []>("op_230"), 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, []>(40861696)))];
            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, []>(40864320)))];
            tensor<fp16, []> var_236_to_fp16 = const()[name = tensor<string, []>("op_236_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_236_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, []>(40866944)))];
            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, []>(44143808)))];
            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, []>(44146432)))];
            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, []>(47423296)))];
            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, []>(50700160)))];
            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, -3, -1])];
            tensor<int32, [4]> transpose_131_perm_0 = const()[name = tensor<string, []>("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_230, 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, []>(50702784)))];
            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, []>(53979648)))];
            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, []>(53982272)))];
            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, []>(53984896)))];
            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_236_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = tensor<string, []>("op_310_cast_fp16")];
            tensor<fp16, [2240, 1280]> var_319_to_fp16 = const()[name = tensor<string, []>("op_319_to_fp16"), val = tensor<fp16, [2240, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53987520)))];
            tensor<fp16, [2240]> var_320_to_fp16 = const()[name = tensor<string, []>("op_320_to_fp16"), val = tensor<fp16, [2240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59721984)))];
            tensor<fp16, [1, 1500, 2240]> 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, 2240]> 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, 2240]> var_325_to_fp16 = const()[name = tensor<string, []>("op_325_to_fp16"), val = tensor<fp16, [1280, 2240]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59726528)))];
            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, []>(65460992)))];
            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_352_axes_0 = const()[name = tensor<string, []>("op_352_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, []>(65463616)))];
            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, []>(65466240)))];
            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_352_cast_fp16 = layer_norm(axes = var_352_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_352_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_363_to_fp16 = const()[name = tensor<string, []>("op_363_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65468864)))];
            tensor<fp16, [1280]> var_364_to_fp16 = const()[name = tensor<string, []>("op_364_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68745728)))];
            tensor<fp16, [1, 1500, 1280]> linear_12_cast_fp16 = linear(bias = var_364_to_fp16, weight = var_363_to_fp16, x = var_352_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_367_to_fp16 = const()[name = tensor<string, []>("op_367_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68748352)))];
            tensor<fp16, [1, 1500, 1280]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_367_to_fp16, x = var_352_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_371_to_fp16 = const()[name = tensor<string, []>("op_371_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72025216)))];
            tensor<fp16, [1280]> var_372_to_fp16 = const()[name = tensor<string, []>("op_372_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75302080)))];
            tensor<fp16, [1, 1500, 1280]> linear_14_cast_fp16 = linear(bias = var_372_to_fp16, weight = var_371_to_fp16, x = var_352_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<int32, [4]> var_380 = const()[name = tensor<string, []>("op_380"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_381_cast_fp16 = reshape(shape = var_380, x = linear_12_cast_fp16)[name = tensor<string, []>("op_381_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_381_cast_fp16, y = const_228_to_fp16)[name = tensor<string, []>("q_11_cast_fp16")];
            tensor<int32, [4]> var_387 = const()[name = tensor<string, []>("op_387"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_388_cast_fp16 = reshape(shape = var_387, x = linear_13_cast_fp16)[name = tensor<string, []>("op_388_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_388_cast_fp16, y = const_229_to_fp16)[name = tensor<string, []>("k_11_cast_fp16")];
            tensor<int32, [4]> var_394 = const()[name = tensor<string, []>("op_394"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_395_cast_fp16 = reshape(shape = var_394, x = linear_14_cast_fp16)[name = tensor<string, []>("op_395_cast_fp16")];
            tensor<int32, [4]> var_396 = const()[name = tensor<string, []>("op_396"), 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, -3, -1])];
            tensor<int32, [4]> transpose_133_perm_0 = const()[name = tensor<string, []>("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_400_cast_fp16 = softmax(axis = var_336, x = qk_5_cast_fp16)[name = tensor<string, []>("op_400_cast_fp16")];
            tensor<bool, []> var_402_transpose_x_0 = const()[name = tensor<string, []>("op_402_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_402_transpose_y_0 = const()[name = tensor<string, []>("op_402_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_311 = transpose(perm = var_396, x = var_395_cast_fp16)[name = tensor<string, []>("transpose_311")];
            tensor<fp16, [1, 20, 1500, 64]> var_402_cast_fp16 = matmul(transpose_x = var_402_transpose_x_0, transpose_y = var_402_transpose_y_0, x = var_400_cast_fp16, y = transpose_311)[name = tensor<string, []>("op_402_cast_fp16")];
            tensor<int32, [4]> var_403 = const()[name = tensor<string, []>("op_403"), 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_403, x = var_402_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_408_to_fp16 = const()[name = tensor<string, []>("op_408_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75304704)))];
            tensor<fp16, [1280]> var_409_to_fp16 = const()[name = tensor<string, []>("op_409_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78581568)))];
            tensor<fp16, [1, 1500, 1280]> linear_15_cast_fp16 = linear(bias = var_409_to_fp16, weight = var_408_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_416_axes_0 = const()[name = tensor<string, []>("op_416_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, []>(78584192)))];
            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, []>(78586816)))];
            tensor<fp16, [1, 1500, 1280]> var_416_cast_fp16 = layer_norm(axes = var_416_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_416_cast_fp16")];
            tensor<fp16, [2048, 1280]> var_425_to_fp16 = const()[name = tensor<string, []>("op_425_to_fp16"), val = tensor<fp16, [2048, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78589440)))];
            tensor<fp16, [2048]> var_426_to_fp16 = const()[name = tensor<string, []>("op_426_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83832384)))];
            tensor<fp16, [1, 1500, 2048]> linear_16_cast_fp16 = linear(bias = var_426_to_fp16, weight = var_425_to_fp16, x = var_416_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, 2048]> 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, 2048]> var_431_to_fp16 = const()[name = tensor<string, []>("op_431_to_fp16"), val = tensor<fp16, [1280, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83836544)))];
            tensor<fp16, [1280]> var_432_to_fp16 = const()[name = tensor<string, []>("op_432_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89079488)))];
            tensor<fp16, [1, 1500, 1280]> linear_17_cast_fp16 = linear(bias = var_432_to_fp16, weight = var_431_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_442 = const()[name = tensor<string, []>("op_442"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_458_axes_0 = const()[name = tensor<string, []>("op_458_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, []>(89082112)))];
            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, []>(89084736)))];
            tensor<fp16, []> var_448_to_fp16 = const()[name = tensor<string, []>("op_448_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_458_cast_fp16 = layer_norm(axes = var_458_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_448_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = tensor<string, []>("op_458_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_469_to_fp16 = const()[name = tensor<string, []>("op_469_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89087360)))];
            tensor<fp16, [1280]> var_470_to_fp16 = const()[name = tensor<string, []>("op_470_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92364224)))];
            tensor<fp16, [1, 1500, 1280]> linear_18_cast_fp16 = linear(bias = var_470_to_fp16, weight = var_469_to_fp16, x = var_458_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_473_to_fp16 = const()[name = tensor<string, []>("op_473_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92366848)))];
            tensor<fp16, [1, 1500, 1280]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_473_to_fp16, x = var_458_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_477_to_fp16 = const()[name = tensor<string, []>("op_477_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95643712)))];
            tensor<fp16, [1280]> var_478_to_fp16 = const()[name = tensor<string, []>("op_478_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98920576)))];
            tensor<fp16, [1, 1500, 1280]> linear_20_cast_fp16 = linear(bias = var_478_to_fp16, weight = var_477_to_fp16, x = var_458_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<int32, [4]> var_486 = const()[name = tensor<string, []>("op_486"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_487_cast_fp16 = reshape(shape = var_486, x = linear_18_cast_fp16)[name = tensor<string, []>("op_487_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_487_cast_fp16, y = const_230_to_fp16)[name = tensor<string, []>("q_15_cast_fp16")];
            tensor<int32, [4]> var_493 = const()[name = tensor<string, []>("op_493"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_494_cast_fp16 = reshape(shape = var_493, x = linear_19_cast_fp16)[name = tensor<string, []>("op_494_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_494_cast_fp16, y = const_231_to_fp16)[name = tensor<string, []>("k_15_cast_fp16")];
            tensor<int32, [4]> var_500 = const()[name = tensor<string, []>("op_500"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_501_cast_fp16 = reshape(shape = var_500, x = linear_20_cast_fp16)[name = tensor<string, []>("op_501_cast_fp16")];
            tensor<int32, [4]> var_502 = const()[name = tensor<string, []>("op_502"), 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, -3, -1])];
            tensor<int32, [4]> transpose_135_perm_0 = const()[name = tensor<string, []>("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_506_cast_fp16 = softmax(axis = var_442, x = qk_7_cast_fp16)[name = tensor<string, []>("op_506_cast_fp16")];
            tensor<bool, []> var_508_transpose_x_0 = const()[name = tensor<string, []>("op_508_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_508_transpose_y_0 = const()[name = tensor<string, []>("op_508_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_307 = transpose(perm = var_502, x = var_501_cast_fp16)[name = tensor<string, []>("transpose_307")];
            tensor<fp16, [1, 20, 1500, 64]> var_508_cast_fp16 = matmul(transpose_x = var_508_transpose_x_0, transpose_y = var_508_transpose_y_0, x = var_506_cast_fp16, y = transpose_307)[name = tensor<string, []>("op_508_cast_fp16")];
            tensor<int32, [4]> var_509 = const()[name = tensor<string, []>("op_509"), 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_509, x = var_508_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_514_to_fp16 = const()[name = tensor<string, []>("op_514_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98923200)))];
            tensor<fp16, [1280]> var_515_to_fp16 = const()[name = tensor<string, []>("op_515_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102200064)))];
            tensor<fp16, [1, 1500, 1280]> linear_21_cast_fp16 = linear(bias = var_515_to_fp16, weight = var_514_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_522_axes_0 = const()[name = tensor<string, []>("op_522_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, []>(102202688)))];
            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, []>(102205312)))];
            tensor<fp16, [1, 1500, 1280]> var_522_cast_fp16 = layer_norm(axes = var_522_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_448_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = tensor<string, []>("op_522_cast_fp16")];
            tensor<fp16, [1824, 1280]> var_531_to_fp16 = const()[name = tensor<string, []>("op_531_to_fp16"), val = tensor<fp16, [1824, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102207936)))];
            tensor<fp16, [1824]> var_532_to_fp16 = const()[name = tensor<string, []>("op_532_to_fp16"), val = tensor<fp16, [1824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106877440)))];
            tensor<fp16, [1, 1500, 1824]> linear_22_cast_fp16 = linear(bias = var_532_to_fp16, weight = var_531_to_fp16, x = var_522_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, 1824]> 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, 1824]> var_537_to_fp16 = const()[name = tensor<string, []>("op_537_to_fp16"), val = tensor<fp16, [1280, 1824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106881152)))];
            tensor<fp16, [1280]> var_538_to_fp16 = const()[name = tensor<string, []>("op_538_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111550656)))];
            tensor<fp16, [1, 1500, 1280]> linear_23_cast_fp16 = linear(bias = var_538_to_fp16, weight = var_537_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_548 = const()[name = tensor<string, []>("op_548"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_564_axes_0 = const()[name = tensor<string, []>("op_564_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, []>(111553280)))];
            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, []>(111555904)))];
            tensor<fp16, []> var_554_to_fp16 = const()[name = tensor<string, []>("op_554_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_564_cast_fp16 = layer_norm(axes = var_564_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_554_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = tensor<string, []>("op_564_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_575_to_fp16 = const()[name = tensor<string, []>("op_575_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111558528)))];
            tensor<fp16, [1280]> var_576_to_fp16 = const()[name = tensor<string, []>("op_576_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114835392)))];
            tensor<fp16, [1, 1500, 1280]> linear_24_cast_fp16 = linear(bias = var_576_to_fp16, weight = var_575_to_fp16, x = var_564_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_579_to_fp16 = const()[name = tensor<string, []>("op_579_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114838016)))];
            tensor<fp16, [1, 1500, 1280]> linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_579_to_fp16, x = var_564_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_583_to_fp16 = const()[name = tensor<string, []>("op_583_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118114880)))];
            tensor<fp16, [1280]> var_584_to_fp16 = const()[name = tensor<string, []>("op_584_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121391744)))];
            tensor<fp16, [1, 1500, 1280]> linear_26_cast_fp16 = linear(bias = var_584_to_fp16, weight = var_583_to_fp16, x = var_564_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<int32, [4]> var_592 = const()[name = tensor<string, []>("op_592"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_593_cast_fp16 = reshape(shape = var_592, x = linear_24_cast_fp16)[name = tensor<string, []>("op_593_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_593_cast_fp16, y = const_232_to_fp16)[name = tensor<string, []>("q_19_cast_fp16")];
            tensor<int32, [4]> var_599 = const()[name = tensor<string, []>("op_599"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_600_cast_fp16 = reshape(shape = var_599, x = linear_25_cast_fp16)[name = tensor<string, []>("op_600_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_600_cast_fp16, y = const_233_to_fp16)[name = tensor<string, []>("k_19_cast_fp16")];
            tensor<int32, [4]> var_606 = const()[name = tensor<string, []>("op_606"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_607_cast_fp16 = reshape(shape = var_606, x = linear_26_cast_fp16)[name = tensor<string, []>("op_607_cast_fp16")];
            tensor<int32, [4]> var_608 = const()[name = tensor<string, []>("op_608"), 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, -3, -1])];
            tensor<int32, [4]> transpose_137_perm_0 = const()[name = tensor<string, []>("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_612_cast_fp16 = softmax(axis = var_548, x = qk_9_cast_fp16)[name = tensor<string, []>("op_612_cast_fp16")];
            tensor<bool, []> var_614_transpose_x_0 = const()[name = tensor<string, []>("op_614_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_614_transpose_y_0 = const()[name = tensor<string, []>("op_614_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_303 = transpose(perm = var_608, x = var_607_cast_fp16)[name = tensor<string, []>("transpose_303")];
            tensor<fp16, [1, 20, 1500, 64]> var_614_cast_fp16 = matmul(transpose_x = var_614_transpose_x_0, transpose_y = var_614_transpose_y_0, x = var_612_cast_fp16, y = transpose_303)[name = tensor<string, []>("op_614_cast_fp16")];
            tensor<int32, [4]> var_615 = const()[name = tensor<string, []>("op_615"), 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_615, x = var_614_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_620_to_fp16 = const()[name = tensor<string, []>("op_620_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121394368)))];
            tensor<fp16, [1280]> var_621_to_fp16 = const()[name = tensor<string, []>("op_621_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124671232)))];
            tensor<fp16, [1, 1500, 1280]> linear_27_cast_fp16 = linear(bias = var_621_to_fp16, weight = var_620_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_628_axes_0 = const()[name = tensor<string, []>("op_628_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, []>(124673856)))];
            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, []>(124676480)))];
            tensor<fp16, [1, 1500, 1280]> var_628_cast_fp16 = layer_norm(axes = var_628_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_554_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = tensor<string, []>("op_628_cast_fp16")];
            tensor<fp16, [2144, 1280]> var_637_to_fp16 = const()[name = tensor<string, []>("op_637_to_fp16"), val = tensor<fp16, [2144, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124679104)))];
            tensor<fp16, [2144]> var_638_to_fp16 = const()[name = tensor<string, []>("op_638_to_fp16"), val = tensor<fp16, [2144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130167808)))];
            tensor<fp16, [1, 1500, 2144]> linear_28_cast_fp16 = linear(bias = var_638_to_fp16, weight = var_637_to_fp16, x = var_628_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, 2144]> 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, 2144]> var_643_to_fp16 = const()[name = tensor<string, []>("op_643_to_fp16"), val = tensor<fp16, [1280, 2144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130172160)))];
            tensor<fp16, [1280]> var_644_to_fp16 = const()[name = tensor<string, []>("op_644_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135660864)))];
            tensor<fp16, [1, 1500, 1280]> linear_29_cast_fp16 = linear(bias = var_644_to_fp16, weight = var_643_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_654 = const()[name = tensor<string, []>("op_654"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_670_axes_0 = const()[name = tensor<string, []>("op_670_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, []>(135663488)))];
            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, []>(135666112)))];
            tensor<fp16, []> var_660_to_fp16 = const()[name = tensor<string, []>("op_660_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_670_cast_fp16 = layer_norm(axes = var_670_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_660_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = tensor<string, []>("op_670_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_681_to_fp16 = const()[name = tensor<string, []>("op_681_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135668736)))];
            tensor<fp16, [1280]> var_682_to_fp16 = const()[name = tensor<string, []>("op_682_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138945600)))];
            tensor<fp16, [1, 1500, 1280]> linear_30_cast_fp16 = linear(bias = var_682_to_fp16, weight = var_681_to_fp16, x = var_670_cast_fp16)[name = tensor<string, []>("linear_30_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, []>(138948224)))];
            tensor<fp16, [1, 1500, 1280]> linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_685_to_fp16, x = var_670_cast_fp16)[name = tensor<string, []>("linear_31_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, []>(142225088)))];
            tensor<fp16, [1280]> var_690_to_fp16 = const()[name = tensor<string, []>("op_690_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145501952)))];
            tensor<fp16, [1, 1500, 1280]> linear_32_cast_fp16 = linear(bias = var_690_to_fp16, weight = var_689_to_fp16, x = var_670_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<int32, [4]> var_698 = const()[name = tensor<string, []>("op_698"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_699_cast_fp16 = reshape(shape = var_698, x = linear_30_cast_fp16)[name = tensor<string, []>("op_699_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_699_cast_fp16, y = const_234_to_fp16)[name = tensor<string, []>("q_23_cast_fp16")];
            tensor<int32, [4]> var_705 = const()[name = tensor<string, []>("op_705"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_706_cast_fp16 = reshape(shape = var_705, x = linear_31_cast_fp16)[name = tensor<string, []>("op_706_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_706_cast_fp16, y = const_235_to_fp16)[name = tensor<string, []>("k_23_cast_fp16")];
            tensor<int32, [4]> var_712 = const()[name = tensor<string, []>("op_712"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_713_cast_fp16 = reshape(shape = var_712, x = linear_32_cast_fp16)[name = tensor<string, []>("op_713_cast_fp16")];
            tensor<int32, [4]> var_714 = const()[name = tensor<string, []>("op_714"), 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, -3, -1])];
            tensor<int32, [4]> transpose_139_perm_0 = const()[name = tensor<string, []>("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_718_cast_fp16 = softmax(axis = var_654, x = qk_11_cast_fp16)[name = tensor<string, []>("op_718_cast_fp16")];
            tensor<bool, []> var_720_transpose_x_0 = const()[name = tensor<string, []>("op_720_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_720_transpose_y_0 = const()[name = tensor<string, []>("op_720_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_299 = transpose(perm = var_714, x = var_713_cast_fp16)[name = tensor<string, []>("transpose_299")];
            tensor<fp16, [1, 20, 1500, 64]> var_720_cast_fp16 = matmul(transpose_x = var_720_transpose_x_0, transpose_y = var_720_transpose_y_0, x = var_718_cast_fp16, y = transpose_299)[name = tensor<string, []>("op_720_cast_fp16")];
            tensor<int32, [4]> var_721 = const()[name = tensor<string, []>("op_721"), 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_721, x = var_720_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_726_to_fp16 = const()[name = tensor<string, []>("op_726_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145504576)))];
            tensor<fp16, [1280]> var_727_to_fp16 = const()[name = tensor<string, []>("op_727_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148781440)))];
            tensor<fp16, [1, 1500, 1280]> linear_33_cast_fp16 = linear(bias = var_727_to_fp16, weight = var_726_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_734_axes_0 = const()[name = tensor<string, []>("op_734_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, []>(148784064)))];
            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, []>(148786688)))];
            tensor<fp16, [1, 1500, 1280]> var_734_cast_fp16 = layer_norm(axes = var_734_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_660_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = tensor<string, []>("op_734_cast_fp16")];
            tensor<fp16, [2144, 1280]> var_743_to_fp16 = const()[name = tensor<string, []>("op_743_to_fp16"), val = tensor<fp16, [2144, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148789312)))];
            tensor<fp16, [2144]> var_744_to_fp16 = const()[name = tensor<string, []>("op_744_to_fp16"), val = tensor<fp16, [2144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154278016)))];
            tensor<fp16, [1, 1500, 2144]> linear_34_cast_fp16 = linear(bias = var_744_to_fp16, weight = var_743_to_fp16, x = var_734_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, 2144]> 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, 2144]> var_749_to_fp16 = const()[name = tensor<string, []>("op_749_to_fp16"), val = tensor<fp16, [1280, 2144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154282368)))];
            tensor<fp16, [1280]> var_750_to_fp16 = const()[name = tensor<string, []>("op_750_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159771072)))];
            tensor<fp16, [1, 1500, 1280]> linear_35_cast_fp16 = linear(bias = var_750_to_fp16, weight = var_749_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_760 = const()[name = tensor<string, []>("op_760"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_776_axes_0 = const()[name = tensor<string, []>("op_776_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, []>(159773696)))];
            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, []>(159776320)))];
            tensor<fp16, []> var_766_to_fp16 = const()[name = tensor<string, []>("op_766_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_776_cast_fp16 = layer_norm(axes = var_776_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_766_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = tensor<string, []>("op_776_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_787_to_fp16 = const()[name = tensor<string, []>("op_787_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159778944)))];
            tensor<fp16, [1280]> var_788_to_fp16 = const()[name = tensor<string, []>("op_788_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163055808)))];
            tensor<fp16, [1, 1500, 1280]> linear_36_cast_fp16 = linear(bias = var_788_to_fp16, weight = var_787_to_fp16, x = var_776_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_791_to_fp16 = const()[name = tensor<string, []>("op_791_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163058432)))];
            tensor<fp16, [1, 1500, 1280]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_791_to_fp16, x = var_776_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_795_to_fp16 = const()[name = tensor<string, []>("op_795_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166335296)))];
            tensor<fp16, [1280]> var_796_to_fp16 = const()[name = tensor<string, []>("op_796_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169612160)))];
            tensor<fp16, [1, 1500, 1280]> linear_38_cast_fp16 = linear(bias = var_796_to_fp16, weight = var_795_to_fp16, x = var_776_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<int32, [4]> var_804 = const()[name = tensor<string, []>("op_804"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_805_cast_fp16 = reshape(shape = var_804, x = linear_36_cast_fp16)[name = tensor<string, []>("op_805_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_805_cast_fp16, y = const_236_to_fp16)[name = tensor<string, []>("q_27_cast_fp16")];
            tensor<int32, [4]> var_811 = const()[name = tensor<string, []>("op_811"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_812_cast_fp16 = reshape(shape = var_811, x = linear_37_cast_fp16)[name = tensor<string, []>("op_812_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_812_cast_fp16, y = const_237_to_fp16)[name = tensor<string, []>("k_27_cast_fp16")];
            tensor<int32, [4]> var_818 = const()[name = tensor<string, []>("op_818"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_819_cast_fp16 = reshape(shape = var_818, x = linear_38_cast_fp16)[name = tensor<string, []>("op_819_cast_fp16")];
            tensor<int32, [4]> var_820 = const()[name = tensor<string, []>("op_820"), 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, -3, -1])];
            tensor<int32, [4]> transpose_141_perm_0 = const()[name = tensor<string, []>("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_824_cast_fp16 = softmax(axis = var_760, x = qk_13_cast_fp16)[name = tensor<string, []>("op_824_cast_fp16")];
            tensor<bool, []> var_826_transpose_x_0 = const()[name = tensor<string, []>("op_826_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_826_transpose_y_0 = const()[name = tensor<string, []>("op_826_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_295 = transpose(perm = var_820, x = var_819_cast_fp16)[name = tensor<string, []>("transpose_295")];
            tensor<fp16, [1, 20, 1500, 64]> var_826_cast_fp16 = matmul(transpose_x = var_826_transpose_x_0, transpose_y = var_826_transpose_y_0, x = var_824_cast_fp16, y = transpose_295)[name = tensor<string, []>("op_826_cast_fp16")];
            tensor<int32, [4]> var_827 = const()[name = tensor<string, []>("op_827"), 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_827, x = var_826_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_832_to_fp16 = const()[name = tensor<string, []>("op_832_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169614784)))];
            tensor<fp16, [1280]> var_833_to_fp16 = const()[name = tensor<string, []>("op_833_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172891648)))];
            tensor<fp16, [1, 1500, 1280]> linear_39_cast_fp16 = linear(bias = var_833_to_fp16, weight = var_832_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_840_axes_0 = const()[name = tensor<string, []>("op_840_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, []>(172894272)))];
            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, []>(172896896)))];
            tensor<fp16, [1, 1500, 1280]> var_840_cast_fp16 = layer_norm(axes = var_840_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_766_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = tensor<string, []>("op_840_cast_fp16")];
            tensor<fp16, [2112, 1280]> var_849_to_fp16 = const()[name = tensor<string, []>("op_849_to_fp16"), val = tensor<fp16, [2112, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172899520)))];
            tensor<fp16, [2112]> var_850_to_fp16 = const()[name = tensor<string, []>("op_850_to_fp16"), val = tensor<fp16, [2112]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178306304)))];
            tensor<fp16, [1, 1500, 2112]> linear_40_cast_fp16 = linear(bias = var_850_to_fp16, weight = var_849_to_fp16, x = var_840_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, 2112]> 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, 2112]> var_855_to_fp16 = const()[name = tensor<string, []>("op_855_to_fp16"), val = tensor<fp16, [1280, 2112]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178310592)))];
            tensor<fp16, [1280]> var_856_to_fp16 = const()[name = tensor<string, []>("op_856_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183717376)))];
            tensor<fp16, [1, 1500, 1280]> linear_41_cast_fp16 = linear(bias = var_856_to_fp16, weight = var_855_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_866 = const()[name = tensor<string, []>("op_866"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_882_axes_0 = const()[name = tensor<string, []>("op_882_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, []>(183720000)))];
            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, []>(183722624)))];
            tensor<fp16, []> var_872_to_fp16 = const()[name = tensor<string, []>("op_872_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_882_cast_fp16 = layer_norm(axes = var_882_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_872_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = tensor<string, []>("op_882_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_893_to_fp16 = const()[name = tensor<string, []>("op_893_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183725248)))];
            tensor<fp16, [1280]> var_894_to_fp16 = const()[name = tensor<string, []>("op_894_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187002112)))];
            tensor<fp16, [1, 1500, 1280]> linear_42_cast_fp16 = linear(bias = var_894_to_fp16, weight = var_893_to_fp16, x = var_882_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_897_to_fp16 = const()[name = tensor<string, []>("op_897_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187004736)))];
            tensor<fp16, [1, 1500, 1280]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_897_to_fp16, x = var_882_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_901_to_fp16 = const()[name = tensor<string, []>("op_901_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190281600)))];
            tensor<fp16, [1280]> var_902_to_fp16 = const()[name = tensor<string, []>("op_902_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193558464)))];
            tensor<fp16, [1, 1500, 1280]> linear_44_cast_fp16 = linear(bias = var_902_to_fp16, weight = var_901_to_fp16, x = var_882_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<int32, [4]> var_910 = const()[name = tensor<string, []>("op_910"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_911_cast_fp16 = reshape(shape = var_910, x = linear_42_cast_fp16)[name = tensor<string, []>("op_911_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_911_cast_fp16, y = const_238_to_fp16)[name = tensor<string, []>("q_31_cast_fp16")];
            tensor<int32, [4]> var_917 = const()[name = tensor<string, []>("op_917"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_918_cast_fp16 = reshape(shape = var_917, x = linear_43_cast_fp16)[name = tensor<string, []>("op_918_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_918_cast_fp16, y = const_239_to_fp16)[name = tensor<string, []>("k_31_cast_fp16")];
            tensor<int32, [4]> var_924 = const()[name = tensor<string, []>("op_924"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_925_cast_fp16 = reshape(shape = var_924, x = linear_44_cast_fp16)[name = tensor<string, []>("op_925_cast_fp16")];
            tensor<int32, [4]> var_926 = const()[name = tensor<string, []>("op_926"), 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, -3, -1])];
            tensor<int32, [4]> transpose_143_perm_0 = const()[name = tensor<string, []>("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_930_cast_fp16 = softmax(axis = var_866, x = qk_15_cast_fp16)[name = tensor<string, []>("op_930_cast_fp16")];
            tensor<bool, []> var_932_transpose_x_0 = const()[name = tensor<string, []>("op_932_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_932_transpose_y_0 = const()[name = tensor<string, []>("op_932_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_291 = transpose(perm = var_926, x = var_925_cast_fp16)[name = tensor<string, []>("transpose_291")];
            tensor<fp16, [1, 20, 1500, 64]> var_932_cast_fp16 = matmul(transpose_x = var_932_transpose_x_0, transpose_y = var_932_transpose_y_0, x = var_930_cast_fp16, y = transpose_291)[name = tensor<string, []>("op_932_cast_fp16")];
            tensor<int32, [4]> var_933 = const()[name = tensor<string, []>("op_933"), 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_933, x = var_932_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_938_to_fp16 = const()[name = tensor<string, []>("op_938_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193561088)))];
            tensor<fp16, [1280]> var_939_to_fp16 = const()[name = tensor<string, []>("op_939_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196837952)))];
            tensor<fp16, [1, 1500, 1280]> linear_45_cast_fp16 = linear(bias = var_939_to_fp16, weight = var_938_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_946_axes_0 = const()[name = tensor<string, []>("op_946_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, []>(196840576)))];
            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, []>(196843200)))];
            tensor<fp16, [1, 1500, 1280]> var_946_cast_fp16 = layer_norm(axes = var_946_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_872_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = tensor<string, []>("op_946_cast_fp16")];
            tensor<fp16, [2112, 1280]> var_955_to_fp16 = const()[name = tensor<string, []>("op_955_to_fp16"), val = tensor<fp16, [2112, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196845824)))];
            tensor<fp16, [2112]> var_956_to_fp16 = const()[name = tensor<string, []>("op_956_to_fp16"), val = tensor<fp16, [2112]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202252608)))];
            tensor<fp16, [1, 1500, 2112]> linear_46_cast_fp16 = linear(bias = var_956_to_fp16, weight = var_955_to_fp16, x = var_946_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, 2112]> 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, 2112]> var_961_to_fp16 = const()[name = tensor<string, []>("op_961_to_fp16"), val = tensor<fp16, [1280, 2112]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202256896)))];
            tensor<fp16, [1280]> var_962_to_fp16 = const()[name = tensor<string, []>("op_962_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207663680)))];
            tensor<fp16, [1, 1500, 1280]> linear_47_cast_fp16 = linear(bias = var_962_to_fp16, weight = var_961_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_972 = const()[name = tensor<string, []>("op_972"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_988_axes_0 = const()[name = tensor<string, []>("op_988_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, []>(207666304)))];
            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, []>(207668928)))];
            tensor<fp16, []> var_978_to_fp16 = const()[name = tensor<string, []>("op_978_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_988_cast_fp16 = layer_norm(axes = var_988_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_978_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = tensor<string, []>("op_988_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_999_to_fp16 = const()[name = tensor<string, []>("op_999_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207671552)))];
            tensor<fp16, [1280]> var_1000_to_fp16 = const()[name = tensor<string, []>("op_1000_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210948416)))];
            tensor<fp16, [1, 1500, 1280]> linear_48_cast_fp16 = linear(bias = var_1000_to_fp16, weight = var_999_to_fp16, x = var_988_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1003_to_fp16 = const()[name = tensor<string, []>("op_1003_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210951040)))];
            tensor<fp16, [1, 1500, 1280]> linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1003_to_fp16, x = var_988_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1007_to_fp16 = const()[name = tensor<string, []>("op_1007_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214227904)))];
            tensor<fp16, [1280]> var_1008_to_fp16 = const()[name = tensor<string, []>("op_1008_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217504768)))];
            tensor<fp16, [1, 1500, 1280]> linear_50_cast_fp16 = linear(bias = var_1008_to_fp16, weight = var_1007_to_fp16, x = var_988_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> var_1016 = const()[name = tensor<string, []>("op_1016"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1017_cast_fp16 = reshape(shape = var_1016, x = linear_48_cast_fp16)[name = tensor<string, []>("op_1017_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_1017_cast_fp16, y = const_240_to_fp16)[name = tensor<string, []>("q_35_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_49_cast_fp16)[name = tensor<string, []>("op_1024_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_1024_cast_fp16, y = const_241_to_fp16)[name = tensor<string, []>("k_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_50_cast_fp16)[name = tensor<string, []>("op_1031_cast_fp16")];
            tensor<int32, [4]> var_1032 = const()[name = tensor<string, []>("op_1032"), 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, -3, -1])];
            tensor<int32, [4]> transpose_145_perm_0 = const()[name = tensor<string, []>("transpose_145_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1036_cast_fp16 = softmax(axis = var_972, x = qk_17_cast_fp16)[name = tensor<string, []>("op_1036_cast_fp16")];
            tensor<bool, []> var_1038_transpose_x_0 = const()[name = tensor<string, []>("op_1038_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1038_transpose_y_0 = const()[name = tensor<string, []>("op_1038_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_287 = transpose(perm = var_1032, x = var_1031_cast_fp16)[name = tensor<string, []>("transpose_287")];
            tensor<fp16, [1, 20, 1500, 64]> var_1038_cast_fp16 = matmul(transpose_x = var_1038_transpose_x_0, transpose_y = var_1038_transpose_y_0, x = var_1036_cast_fp16, y = transpose_287)[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<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_1039, x = var_1038_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_1044_to_fp16 = const()[name = tensor<string, []>("op_1044_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217507392)))];
            tensor<fp16, [1280]> var_1045_to_fp16 = const()[name = tensor<string, []>("op_1045_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220784256)))];
            tensor<fp16, [1, 1500, 1280]> linear_51_cast_fp16 = linear(bias = var_1045_to_fp16, weight = var_1044_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_1052_axes_0 = const()[name = tensor<string, []>("op_1052_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, []>(220786880)))];
            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, []>(220789504)))];
            tensor<fp16, [1, 1500, 1280]> var_1052_cast_fp16 = layer_norm(axes = var_1052_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_978_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = tensor<string, []>("op_1052_cast_fp16")];
            tensor<fp16, [2304, 1280]> var_1061_to_fp16 = const()[name = tensor<string, []>("op_1061_to_fp16"), val = tensor<fp16, [2304, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220792128)))];
            tensor<fp16, [2304]> var_1062_to_fp16 = const()[name = tensor<string, []>("op_1062_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226690432)))];
            tensor<fp16, [1, 1500, 2304]> linear_52_cast_fp16 = linear(bias = var_1062_to_fp16, weight = var_1061_to_fp16, x = var_1052_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, 2304]> 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, 2304]> var_1067_to_fp16 = const()[name = tensor<string, []>("op_1067_to_fp16"), val = tensor<fp16, [1280, 2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226695104)))];
            tensor<fp16, [1280]> var_1068_to_fp16 = const()[name = tensor<string, []>("op_1068_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(232593408)))];
            tensor<fp16, [1, 1500, 1280]> linear_53_cast_fp16 = linear(bias = var_1068_to_fp16, weight = var_1067_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_1078 = const()[name = tensor<string, []>("op_1078"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1094_axes_0 = const()[name = tensor<string, []>("op_1094_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, []>(232596032)))];
            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, []>(232598656)))];
            tensor<fp16, []> var_1084_to_fp16 = const()[name = tensor<string, []>("op_1084_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1094_cast_fp16 = layer_norm(axes = var_1094_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1084_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = tensor<string, []>("op_1094_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1105_to_fp16 = const()[name = tensor<string, []>("op_1105_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(232601280)))];
            tensor<fp16, [1280]> var_1106_to_fp16 = const()[name = tensor<string, []>("op_1106_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235878144)))];
            tensor<fp16, [1, 1500, 1280]> linear_54_cast_fp16 = linear(bias = var_1106_to_fp16, weight = var_1105_to_fp16, x = var_1094_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1109_to_fp16 = const()[name = tensor<string, []>("op_1109_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235880768)))];
            tensor<fp16, [1, 1500, 1280]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1109_to_fp16, x = var_1094_cast_fp16)[name = tensor<string, []>("linear_55_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, []>(239157632)))];
            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, []>(242434496)))];
            tensor<fp16, [1, 1500, 1280]> linear_56_cast_fp16 = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1094_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<int32, [4]> var_1122 = const()[name = tensor<string, []>("op_1122"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1123_cast_fp16 = reshape(shape = var_1122, x = linear_54_cast_fp16)[name = tensor<string, []>("op_1123_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_1123_cast_fp16, y = const_242_to_fp16)[name = tensor<string, []>("q_39_cast_fp16")];
            tensor<int32, [4]> var_1129 = const()[name = tensor<string, []>("op_1129"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1130_cast_fp16 = reshape(shape = var_1129, x = linear_55_cast_fp16)[name = tensor<string, []>("op_1130_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_1130_cast_fp16, y = const_243_to_fp16)[name = tensor<string, []>("k_39_cast_fp16")];
            tensor<int32, [4]> var_1136 = const()[name = tensor<string, []>("op_1136"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1137_cast_fp16 = reshape(shape = var_1136, x = linear_56_cast_fp16)[name = tensor<string, []>("op_1137_cast_fp16")];
            tensor<int32, [4]> var_1138 = const()[name = tensor<string, []>("op_1138"), 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, -3, -1])];
            tensor<int32, [4]> transpose_147_perm_0 = const()[name = tensor<string, []>("transpose_147_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1142_cast_fp16 = softmax(axis = var_1078, x = qk_19_cast_fp16)[name = tensor<string, []>("op_1142_cast_fp16")];
            tensor<bool, []> var_1144_transpose_x_0 = const()[name = tensor<string, []>("op_1144_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1144_transpose_y_0 = const()[name = tensor<string, []>("op_1144_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_283 = transpose(perm = var_1138, x = var_1137_cast_fp16)[name = tensor<string, []>("transpose_283")];
            tensor<fp16, [1, 20, 1500, 64]> var_1144_cast_fp16 = matmul(transpose_x = var_1144_transpose_x_0, transpose_y = var_1144_transpose_y_0, x = var_1142_cast_fp16, y = transpose_283)[name = tensor<string, []>("op_1144_cast_fp16")];
            tensor<int32, [4]> var_1145 = const()[name = tensor<string, []>("op_1145"), 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_1145, x = var_1144_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_1150_to_fp16 = const()[name = tensor<string, []>("op_1150_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(242437120)))];
            tensor<fp16, [1280]> var_1151_to_fp16 = const()[name = tensor<string, []>("op_1151_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245713984)))];
            tensor<fp16, [1, 1500, 1280]> linear_57_cast_fp16 = linear(bias = var_1151_to_fp16, weight = var_1150_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_1158_axes_0 = const()[name = tensor<string, []>("op_1158_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, []>(245716608)))];
            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, []>(245719232)))];
            tensor<fp16, [1, 1500, 1280]> var_1158_cast_fp16 = layer_norm(axes = var_1158_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1084_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = tensor<string, []>("op_1158_cast_fp16")];
            tensor<fp16, [2624, 1280]> var_1167_to_fp16 = const()[name = tensor<string, []>("op_1167_to_fp16"), val = tensor<fp16, [2624, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245721856)))];
            tensor<fp16, [2624]> var_1168_to_fp16 = const()[name = tensor<string, []>("op_1168_to_fp16"), val = tensor<fp16, [2624]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252439360)))];
            tensor<fp16, [1, 1500, 2624]> linear_58_cast_fp16 = linear(bias = var_1168_to_fp16, weight = var_1167_to_fp16, x = var_1158_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, 2624]> 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, 2624]> var_1173_to_fp16 = const()[name = tensor<string, []>("op_1173_to_fp16"), val = tensor<fp16, [1280, 2624]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252444672)))];
            tensor<fp16, [1280]> var_1174_to_fp16 = const()[name = tensor<string, []>("op_1174_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259162176)))];
            tensor<fp16, [1, 1500, 1280]> linear_59_cast_fp16 = linear(bias = var_1174_to_fp16, weight = var_1173_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_1184 = const()[name = tensor<string, []>("op_1184"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1200_axes_0 = const()[name = tensor<string, []>("op_1200_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, []>(259164800)))];
            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, []>(259167424)))];
            tensor<fp16, []> var_1190_to_fp16 = const()[name = tensor<string, []>("op_1190_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1200_cast_fp16 = layer_norm(axes = var_1200_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1190_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = tensor<string, []>("op_1200_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1211_to_fp16 = const()[name = tensor<string, []>("op_1211_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259170048)))];
            tensor<fp16, [1280]> var_1212_to_fp16 = const()[name = tensor<string, []>("op_1212_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262446912)))];
            tensor<fp16, [1, 1500, 1280]> linear_60_cast_fp16 = linear(bias = var_1212_to_fp16, weight = var_1211_to_fp16, x = var_1200_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1215_to_fp16 = const()[name = tensor<string, []>("op_1215_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262449536)))];
            tensor<fp16, [1, 1500, 1280]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1215_to_fp16, x = var_1200_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1219_to_fp16 = const()[name = tensor<string, []>("op_1219_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265726400)))];
            tensor<fp16, [1280]> var_1220_to_fp16 = const()[name = tensor<string, []>("op_1220_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269003264)))];
            tensor<fp16, [1, 1500, 1280]> linear_62_cast_fp16 = linear(bias = var_1220_to_fp16, weight = var_1219_to_fp16, x = var_1200_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<int32, [4]> var_1228 = const()[name = tensor<string, []>("op_1228"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1229_cast_fp16 = reshape(shape = var_1228, x = linear_60_cast_fp16)[name = tensor<string, []>("op_1229_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_1229_cast_fp16, y = const_244_to_fp16)[name = tensor<string, []>("q_43_cast_fp16")];
            tensor<int32, [4]> var_1235 = const()[name = tensor<string, []>("op_1235"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1236_cast_fp16 = reshape(shape = var_1235, x = linear_61_cast_fp16)[name = tensor<string, []>("op_1236_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_1236_cast_fp16, y = const_245_to_fp16)[name = tensor<string, []>("k_43_cast_fp16")];
            tensor<int32, [4]> var_1242 = const()[name = tensor<string, []>("op_1242"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1243_cast_fp16 = reshape(shape = var_1242, x = linear_62_cast_fp16)[name = tensor<string, []>("op_1243_cast_fp16")];
            tensor<int32, [4]> var_1244 = const()[name = tensor<string, []>("op_1244"), 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, -3, -1])];
            tensor<int32, [4]> transpose_149_perm_0 = const()[name = tensor<string, []>("transpose_149_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1248_cast_fp16 = softmax(axis = var_1184, x = qk_21_cast_fp16)[name = tensor<string, []>("op_1248_cast_fp16")];
            tensor<bool, []> var_1250_transpose_x_0 = const()[name = tensor<string, []>("op_1250_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1250_transpose_y_0 = const()[name = tensor<string, []>("op_1250_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_279 = transpose(perm = var_1244, x = var_1243_cast_fp16)[name = tensor<string, []>("transpose_279")];
            tensor<fp16, [1, 20, 1500, 64]> var_1250_cast_fp16 = matmul(transpose_x = var_1250_transpose_x_0, transpose_y = var_1250_transpose_y_0, x = var_1248_cast_fp16, y = transpose_279)[name = tensor<string, []>("op_1250_cast_fp16")];
            tensor<int32, [4]> var_1251 = const()[name = tensor<string, []>("op_1251"), 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_1251, x = var_1250_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_1256_to_fp16 = const()[name = tensor<string, []>("op_1256_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269005888)))];
            tensor<fp16, [1280]> var_1257_to_fp16 = const()[name = tensor<string, []>("op_1257_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(272282752)))];
            tensor<fp16, [1, 1500, 1280]> linear_63_cast_fp16 = linear(bias = var_1257_to_fp16, weight = var_1256_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_1264_axes_0 = const()[name = tensor<string, []>("op_1264_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, []>(272285376)))];
            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, []>(272288000)))];
            tensor<fp16, [1, 1500, 1280]> var_1264_cast_fp16 = layer_norm(axes = var_1264_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1190_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = tensor<string, []>("op_1264_cast_fp16")];
            tensor<fp16, [3072, 1280]> var_1273_to_fp16 = const()[name = tensor<string, []>("op_1273_to_fp16"), val = tensor<fp16, [3072, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(272290624)))];
            tensor<fp16, [3072]> var_1274_to_fp16 = const()[name = tensor<string, []>("op_1274_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(280155008)))];
            tensor<fp16, [1, 1500, 3072]> linear_64_cast_fp16 = linear(bias = var_1274_to_fp16, weight = var_1273_to_fp16, x = var_1264_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, 3072]> 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, 3072]> var_1279_to_fp16 = const()[name = tensor<string, []>("op_1279_to_fp16"), val = tensor<fp16, [1280, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(280161216)))];
            tensor<fp16, [1280]> var_1280_to_fp16 = const()[name = tensor<string, []>("op_1280_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288025600)))];
            tensor<fp16, [1, 1500, 1280]> linear_65_cast_fp16 = linear(bias = var_1280_to_fp16, weight = var_1279_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_1290 = const()[name = tensor<string, []>("op_1290"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1306_axes_0 = const()[name = tensor<string, []>("op_1306_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, []>(288028224)))];
            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, []>(288030848)))];
            tensor<fp16, []> var_1296_to_fp16 = const()[name = tensor<string, []>("op_1296_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1306_cast_fp16 = layer_norm(axes = var_1306_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1296_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = tensor<string, []>("op_1306_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1317_to_fp16 = const()[name = tensor<string, []>("op_1317_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288033472)))];
            tensor<fp16, [1280]> var_1318_to_fp16 = const()[name = tensor<string, []>("op_1318_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291310336)))];
            tensor<fp16, [1, 1500, 1280]> linear_66_cast_fp16 = linear(bias = var_1318_to_fp16, weight = var_1317_to_fp16, x = var_1306_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1321_to_fp16 = const()[name = tensor<string, []>("op_1321_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291312960)))];
            tensor<fp16, [1, 1500, 1280]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1321_to_fp16, x = var_1306_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1325_to_fp16 = const()[name = tensor<string, []>("op_1325_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(294589824)))];
            tensor<fp16, [1280]> var_1326_to_fp16 = const()[name = tensor<string, []>("op_1326_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297866688)))];
            tensor<fp16, [1, 1500, 1280]> linear_68_cast_fp16 = linear(bias = var_1326_to_fp16, weight = var_1325_to_fp16, x = var_1306_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<int32, [4]> var_1334 = const()[name = tensor<string, []>("op_1334"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1335_cast_fp16 = reshape(shape = var_1334, x = linear_66_cast_fp16)[name = tensor<string, []>("op_1335_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_1335_cast_fp16, y = const_246_to_fp16)[name = tensor<string, []>("q_47_cast_fp16")];
            tensor<int32, [4]> var_1341 = const()[name = tensor<string, []>("op_1341"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1342_cast_fp16 = reshape(shape = var_1341, x = linear_67_cast_fp16)[name = tensor<string, []>("op_1342_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_1342_cast_fp16, y = const_247_to_fp16)[name = tensor<string, []>("k_47_cast_fp16")];
            tensor<int32, [4]> var_1348 = const()[name = tensor<string, []>("op_1348"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1349_cast_fp16 = reshape(shape = var_1348, x = linear_68_cast_fp16)[name = tensor<string, []>("op_1349_cast_fp16")];
            tensor<int32, [4]> var_1350 = const()[name = tensor<string, []>("op_1350"), 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, -3, -1])];
            tensor<int32, [4]> transpose_151_perm_0 = const()[name = tensor<string, []>("transpose_151_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1354_cast_fp16 = softmax(axis = var_1290, x = qk_23_cast_fp16)[name = tensor<string, []>("op_1354_cast_fp16")];
            tensor<bool, []> var_1356_transpose_x_0 = const()[name = tensor<string, []>("op_1356_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1356_transpose_y_0 = const()[name = tensor<string, []>("op_1356_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_275 = transpose(perm = var_1350, x = var_1349_cast_fp16)[name = tensor<string, []>("transpose_275")];
            tensor<fp16, [1, 20, 1500, 64]> var_1356_cast_fp16 = matmul(transpose_x = var_1356_transpose_x_0, transpose_y = var_1356_transpose_y_0, x = var_1354_cast_fp16, y = transpose_275)[name = tensor<string, []>("op_1356_cast_fp16")];
            tensor<int32, [4]> var_1357 = const()[name = tensor<string, []>("op_1357"), 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_1357, x = var_1356_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_1362_to_fp16 = const()[name = tensor<string, []>("op_1362_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297869312)))];
            tensor<fp16, [1280]> var_1363_to_fp16 = const()[name = tensor<string, []>("op_1363_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301146176)))];
            tensor<fp16, [1, 1500, 1280]> linear_69_cast_fp16 = linear(bias = var_1363_to_fp16, weight = var_1362_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_1370_axes_0 = const()[name = tensor<string, []>("op_1370_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, []>(301148800)))];
            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, []>(301151424)))];
            tensor<fp16, [1, 1500, 1280]> var_1370_cast_fp16 = layer_norm(axes = var_1370_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1296_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = tensor<string, []>("op_1370_cast_fp16")];
            tensor<fp16, [2976, 1280]> var_1379_to_fp16 = const()[name = tensor<string, []>("op_1379_to_fp16"), val = tensor<fp16, [2976, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301154048)))];
            tensor<fp16, [2976]> var_1380_to_fp16 = const()[name = tensor<string, []>("op_1380_to_fp16"), val = tensor<fp16, [2976]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(308772672)))];
            tensor<fp16, [1, 1500, 2976]> linear_70_cast_fp16 = linear(bias = var_1380_to_fp16, weight = var_1379_to_fp16, x = var_1370_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, 2976]> 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, 2976]> var_1385_to_fp16 = const()[name = tensor<string, []>("op_1385_to_fp16"), val = tensor<fp16, [1280, 2976]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(308778688)))];
            tensor<fp16, [1280]> var_1386_to_fp16 = const()[name = tensor<string, []>("op_1386_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316397312)))];
            tensor<fp16, [1, 1500, 1280]> linear_71_cast_fp16 = linear(bias = var_1386_to_fp16, weight = var_1385_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_1396 = const()[name = tensor<string, []>("op_1396"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1412_axes_0 = const()[name = tensor<string, []>("op_1412_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, []>(316399936)))];
            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, []>(316402560)))];
            tensor<fp16, []> var_1402_to_fp16 = const()[name = tensor<string, []>("op_1402_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1412_cast_fp16 = layer_norm(axes = var_1412_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1402_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = tensor<string, []>("op_1412_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1423_to_fp16 = const()[name = tensor<string, []>("op_1423_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316405184)))];
            tensor<fp16, [1280]> var_1424_to_fp16 = const()[name = tensor<string, []>("op_1424_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(319682048)))];
            tensor<fp16, [1, 1500, 1280]> linear_72_cast_fp16 = linear(bias = var_1424_to_fp16, weight = var_1423_to_fp16, x = var_1412_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1427_to_fp16 = const()[name = tensor<string, []>("op_1427_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(319684672)))];
            tensor<fp16, [1, 1500, 1280]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1427_to_fp16, x = var_1412_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1431_to_fp16 = const()[name = tensor<string, []>("op_1431_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(322961536)))];
            tensor<fp16, [1280]> var_1432_to_fp16 = const()[name = tensor<string, []>("op_1432_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(326238400)))];
            tensor<fp16, [1, 1500, 1280]> linear_74_cast_fp16 = linear(bias = var_1432_to_fp16, weight = var_1431_to_fp16, x = var_1412_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
            tensor<int32, [4]> var_1440 = const()[name = tensor<string, []>("op_1440"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1441_cast_fp16 = reshape(shape = var_1440, x = linear_72_cast_fp16)[name = tensor<string, []>("op_1441_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_1441_cast_fp16, y = const_248_to_fp16)[name = tensor<string, []>("q_51_cast_fp16")];
            tensor<int32, [4]> var_1447 = const()[name = tensor<string, []>("op_1447"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1448_cast_fp16 = reshape(shape = var_1447, x = linear_73_cast_fp16)[name = tensor<string, []>("op_1448_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_1448_cast_fp16, y = const_249_to_fp16)[name = tensor<string, []>("k_51_cast_fp16")];
            tensor<int32, [4]> var_1454 = const()[name = tensor<string, []>("op_1454"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1455_cast_fp16 = reshape(shape = var_1454, x = linear_74_cast_fp16)[name = tensor<string, []>("op_1455_cast_fp16")];
            tensor<int32, [4]> var_1456 = const()[name = tensor<string, []>("op_1456"), 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, -3, -1])];
            tensor<int32, [4]> transpose_153_perm_0 = const()[name = tensor<string, []>("transpose_153_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1460_cast_fp16 = softmax(axis = var_1396, x = qk_25_cast_fp16)[name = tensor<string, []>("op_1460_cast_fp16")];
            tensor<bool, []> var_1462_transpose_x_0 = const()[name = tensor<string, []>("op_1462_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1462_transpose_y_0 = const()[name = tensor<string, []>("op_1462_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_271 = transpose(perm = var_1456, x = var_1455_cast_fp16)[name = tensor<string, []>("transpose_271")];
            tensor<fp16, [1, 20, 1500, 64]> var_1462_cast_fp16 = matmul(transpose_x = var_1462_transpose_x_0, transpose_y = var_1462_transpose_y_0, x = var_1460_cast_fp16, y = transpose_271)[name = tensor<string, []>("op_1462_cast_fp16")];
            tensor<int32, [4]> var_1463 = const()[name = tensor<string, []>("op_1463"), 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_1463, x = var_1462_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_1468_to_fp16 = const()[name = tensor<string, []>("op_1468_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(326241024)))];
            tensor<fp16, [1280]> var_1469_to_fp16 = const()[name = tensor<string, []>("op_1469_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329517888)))];
            tensor<fp16, [1, 1500, 1280]> linear_75_cast_fp16 = linear(bias = var_1469_to_fp16, weight = var_1468_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_1476_axes_0 = const()[name = tensor<string, []>("op_1476_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, []>(329520512)))];
            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, []>(329523136)))];
            tensor<fp16, [1, 1500, 1280]> var_1476_cast_fp16 = layer_norm(axes = var_1476_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1402_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = tensor<string, []>("op_1476_cast_fp16")];
            tensor<fp16, [3712, 1280]> var_1485_to_fp16 = const()[name = tensor<string, []>("op_1485_to_fp16"), val = tensor<fp16, [3712, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329525760)))];
            tensor<fp16, [3712]> var_1486_to_fp16 = const()[name = tensor<string, []>("op_1486_to_fp16"), val = tensor<fp16, [3712]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339028544)))];
            tensor<fp16, [1, 1500, 3712]> linear_76_cast_fp16 = linear(bias = var_1486_to_fp16, weight = var_1485_to_fp16, x = var_1476_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, 3712]> 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, 3712]> var_1491_to_fp16 = const()[name = tensor<string, []>("op_1491_to_fp16"), val = tensor<fp16, [1280, 3712]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(339036032)))];
            tensor<fp16, [1280]> var_1492_to_fp16 = const()[name = tensor<string, []>("op_1492_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(348538816)))];
            tensor<fp16, [1, 1500, 1280]> linear_77_cast_fp16 = linear(bias = var_1492_to_fp16, weight = var_1491_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_1502 = const()[name = tensor<string, []>("op_1502"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1518_axes_0 = const()[name = tensor<string, []>("op_1518_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, []>(348541440)))];
            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, []>(348544064)))];
            tensor<fp16, []> var_1508_to_fp16 = const()[name = tensor<string, []>("op_1508_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1518_cast_fp16 = layer_norm(axes = var_1518_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1508_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = tensor<string, []>("op_1518_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1529_to_fp16 = const()[name = tensor<string, []>("op_1529_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(348546688)))];
            tensor<fp16, [1280]> var_1530_to_fp16 = const()[name = tensor<string, []>("op_1530_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(351823552)))];
            tensor<fp16, [1, 1500, 1280]> linear_78_cast_fp16 = linear(bias = var_1530_to_fp16, weight = var_1529_to_fp16, x = var_1518_cast_fp16)[name = tensor<string, []>("linear_78_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1533_to_fp16 = const()[name = tensor<string, []>("op_1533_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(351826176)))];
            tensor<fp16, [1, 1500, 1280]> linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1533_to_fp16, x = var_1518_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1537_to_fp16 = const()[name = tensor<string, []>("op_1537_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355103040)))];
            tensor<fp16, [1280]> var_1538_to_fp16 = const()[name = tensor<string, []>("op_1538_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358379904)))];
            tensor<fp16, [1, 1500, 1280]> linear_80_cast_fp16 = linear(bias = var_1538_to_fp16, weight = var_1537_to_fp16, x = var_1518_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")];
            tensor<int32, [4]> var_1546 = const()[name = tensor<string, []>("op_1546"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1547_cast_fp16 = reshape(shape = var_1546, x = linear_78_cast_fp16)[name = tensor<string, []>("op_1547_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_1547_cast_fp16, y = const_250_to_fp16)[name = tensor<string, []>("q_55_cast_fp16")];
            tensor<int32, [4]> var_1553 = const()[name = tensor<string, []>("op_1553"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1554_cast_fp16 = reshape(shape = var_1553, x = linear_79_cast_fp16)[name = tensor<string, []>("op_1554_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_1554_cast_fp16, y = const_251_to_fp16)[name = tensor<string, []>("k_55_cast_fp16")];
            tensor<int32, [4]> var_1560 = const()[name = tensor<string, []>("op_1560"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1561_cast_fp16 = reshape(shape = var_1560, x = linear_80_cast_fp16)[name = tensor<string, []>("op_1561_cast_fp16")];
            tensor<int32, [4]> var_1562 = const()[name = tensor<string, []>("op_1562"), 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, -3, -1])];
            tensor<int32, [4]> transpose_155_perm_0 = const()[name = tensor<string, []>("transpose_155_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1566_cast_fp16 = softmax(axis = var_1502, x = qk_27_cast_fp16)[name = tensor<string, []>("op_1566_cast_fp16")];
            tensor<bool, []> var_1568_transpose_x_0 = const()[name = tensor<string, []>("op_1568_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1568_transpose_y_0 = const()[name = tensor<string, []>("op_1568_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_267 = transpose(perm = var_1562, x = var_1561_cast_fp16)[name = tensor<string, []>("transpose_267")];
            tensor<fp16, [1, 20, 1500, 64]> var_1568_cast_fp16 = matmul(transpose_x = var_1568_transpose_x_0, transpose_y = var_1568_transpose_y_0, x = var_1566_cast_fp16, y = transpose_267)[name = tensor<string, []>("op_1568_cast_fp16")];
            tensor<int32, [4]> var_1569 = const()[name = tensor<string, []>("op_1569"), 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_1569, x = var_1568_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_1574_to_fp16 = const()[name = tensor<string, []>("op_1574_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358382528)))];
            tensor<fp16, [1280]> var_1575_to_fp16 = const()[name = tensor<string, []>("op_1575_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361659392)))];
            tensor<fp16, [1, 1500, 1280]> linear_81_cast_fp16 = linear(bias = var_1575_to_fp16, weight = var_1574_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_1582_axes_0 = const()[name = tensor<string, []>("op_1582_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, []>(361662016)))];
            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, []>(361664640)))];
            tensor<fp16, [1, 1500, 1280]> var_1582_cast_fp16 = layer_norm(axes = var_1582_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1508_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = tensor<string, []>("op_1582_cast_fp16")];
            tensor<fp16, [3872, 1280]> var_1591_to_fp16 = const()[name = tensor<string, []>("op_1591_to_fp16"), val = tensor<fp16, [3872, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(361667264)))];
            tensor<fp16, [3872]> var_1592_to_fp16 = const()[name = tensor<string, []>("op_1592_to_fp16"), val = tensor<fp16, [3872]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(371579648)))];
            tensor<fp16, [1, 1500, 3872]> linear_82_cast_fp16 = linear(bias = var_1592_to_fp16, weight = var_1591_to_fp16, x = var_1582_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, 3872]> 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, 3872]> var_1597_to_fp16 = const()[name = tensor<string, []>("op_1597_to_fp16"), val = tensor<fp16, [1280, 3872]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(371587456)))];
            tensor<fp16, [1280]> var_1598_to_fp16 = const()[name = tensor<string, []>("op_1598_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381499840)))];
            tensor<fp16, [1, 1500, 1280]> linear_83_cast_fp16 = linear(bias = var_1598_to_fp16, weight = var_1597_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_1608 = const()[name = tensor<string, []>("op_1608"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1624_axes_0 = const()[name = tensor<string, []>("op_1624_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, []>(381502464)))];
            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, []>(381505088)))];
            tensor<fp16, []> var_1614_to_fp16 = const()[name = tensor<string, []>("op_1614_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1624_cast_fp16 = layer_norm(axes = var_1624_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1614_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = tensor<string, []>("op_1624_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1635_to_fp16 = const()[name = tensor<string, []>("op_1635_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381507712)))];
            tensor<fp16, [1280]> var_1636_to_fp16 = const()[name = tensor<string, []>("op_1636_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(384784576)))];
            tensor<fp16, [1, 1500, 1280]> linear_84_cast_fp16 = linear(bias = var_1636_to_fp16, weight = var_1635_to_fp16, x = var_1624_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1639_to_fp16 = const()[name = tensor<string, []>("op_1639_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(384787200)))];
            tensor<fp16, [1, 1500, 1280]> linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1639_to_fp16, x = var_1624_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1643_to_fp16 = const()[name = tensor<string, []>("op_1643_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(388064064)))];
            tensor<fp16, [1280]> var_1644_to_fp16 = const()[name = tensor<string, []>("op_1644_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(391340928)))];
            tensor<fp16, [1, 1500, 1280]> linear_86_cast_fp16 = linear(bias = var_1644_to_fp16, weight = var_1643_to_fp16, x = var_1624_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")];
            tensor<int32, [4]> var_1652 = const()[name = tensor<string, []>("op_1652"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1653_cast_fp16 = reshape(shape = var_1652, x = linear_84_cast_fp16)[name = tensor<string, []>("op_1653_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_1653_cast_fp16, y = const_252_to_fp16)[name = tensor<string, []>("q_59_cast_fp16")];
            tensor<int32, [4]> var_1659 = const()[name = tensor<string, []>("op_1659"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1660_cast_fp16 = reshape(shape = var_1659, x = linear_85_cast_fp16)[name = tensor<string, []>("op_1660_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_1660_cast_fp16, y = const_253_to_fp16)[name = tensor<string, []>("k_59_cast_fp16")];
            tensor<int32, [4]> var_1666 = const()[name = tensor<string, []>("op_1666"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1667_cast_fp16 = reshape(shape = var_1666, x = linear_86_cast_fp16)[name = tensor<string, []>("op_1667_cast_fp16")];
            tensor<int32, [4]> var_1668 = const()[name = tensor<string, []>("op_1668"), 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, -3, -1])];
            tensor<int32, [4]> transpose_157_perm_0 = const()[name = tensor<string, []>("transpose_157_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1672_cast_fp16 = softmax(axis = var_1608, x = qk_29_cast_fp16)[name = tensor<string, []>("op_1672_cast_fp16")];
            tensor<bool, []> var_1674_transpose_x_0 = const()[name = tensor<string, []>("op_1674_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1674_transpose_y_0 = const()[name = tensor<string, []>("op_1674_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_263 = transpose(perm = var_1668, x = var_1667_cast_fp16)[name = tensor<string, []>("transpose_263")];
            tensor<fp16, [1, 20, 1500, 64]> var_1674_cast_fp16 = matmul(transpose_x = var_1674_transpose_x_0, transpose_y = var_1674_transpose_y_0, x = var_1672_cast_fp16, y = transpose_263)[name = tensor<string, []>("op_1674_cast_fp16")];
            tensor<int32, [4]> var_1675 = const()[name = tensor<string, []>("op_1675"), 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_1675, x = var_1674_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_1680_to_fp16 = const()[name = tensor<string, []>("op_1680_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(391343552)))];
            tensor<fp16, [1280]> var_1681_to_fp16 = const()[name = tensor<string, []>("op_1681_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(394620416)))];
            tensor<fp16, [1, 1500, 1280]> linear_87_cast_fp16 = linear(bias = var_1681_to_fp16, weight = var_1680_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_1688_axes_0 = const()[name = tensor<string, []>("op_1688_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, []>(394623040)))];
            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, []>(394625664)))];
            tensor<fp16, [1, 1500, 1280]> var_1688_cast_fp16 = layer_norm(axes = var_1688_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1614_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = tensor<string, []>("op_1688_cast_fp16")];
            tensor<fp16, [4128, 1280]> var_1697_to_fp16 = const()[name = tensor<string, []>("op_1697_to_fp16"), val = tensor<fp16, [4128, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(394628288)))];
            tensor<fp16, [4128]> var_1698_to_fp16 = const()[name = tensor<string, []>("op_1698_to_fp16"), val = tensor<fp16, [4128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(405196032)))];
            tensor<fp16, [1, 1500, 4128]> linear_88_cast_fp16 = linear(bias = var_1698_to_fp16, weight = var_1697_to_fp16, x = var_1688_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, 4128]> 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, 4128]> var_1703_to_fp16 = const()[name = tensor<string, []>("op_1703_to_fp16"), val = tensor<fp16, [1280, 4128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(405204352)))];
            tensor<fp16, [1280]> var_1704_to_fp16 = const()[name = tensor<string, []>("op_1704_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415772096)))];
            tensor<fp16, [1, 1500, 1280]> linear_89_cast_fp16 = linear(bias = var_1704_to_fp16, weight = var_1703_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_1714 = const()[name = tensor<string, []>("op_1714"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1730_axes_0 = const()[name = tensor<string, []>("op_1730_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, []>(415774720)))];
            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, []>(415777344)))];
            tensor<fp16, []> var_1720_to_fp16 = const()[name = tensor<string, []>("op_1720_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1730_cast_fp16 = layer_norm(axes = var_1730_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1720_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = tensor<string, []>("op_1730_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1741_to_fp16 = const()[name = tensor<string, []>("op_1741_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415779968)))];
            tensor<fp16, [1280]> var_1742_to_fp16 = const()[name = tensor<string, []>("op_1742_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419056832)))];
            tensor<fp16, [1, 1500, 1280]> linear_90_cast_fp16 = linear(bias = var_1742_to_fp16, weight = var_1741_to_fp16, x = var_1730_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1745_to_fp16 = const()[name = tensor<string, []>("op_1745_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419059456)))];
            tensor<fp16, [1, 1500, 1280]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1745_to_fp16, x = var_1730_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1749_to_fp16 = const()[name = tensor<string, []>("op_1749_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(422336320)))];
            tensor<fp16, [1280]> var_1750_to_fp16 = const()[name = tensor<string, []>("op_1750_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(425613184)))];
            tensor<fp16, [1, 1500, 1280]> linear_92_cast_fp16 = linear(bias = var_1750_to_fp16, weight = var_1749_to_fp16, x = var_1730_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")];
            tensor<int32, [4]> var_1758 = const()[name = tensor<string, []>("op_1758"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1759_cast_fp16 = reshape(shape = var_1758, x = linear_90_cast_fp16)[name = tensor<string, []>("op_1759_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_1759_cast_fp16, y = const_254_to_fp16)[name = tensor<string, []>("q_63_cast_fp16")];
            tensor<int32, [4]> var_1765 = const()[name = tensor<string, []>("op_1765"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1766_cast_fp16 = reshape(shape = var_1765, x = linear_91_cast_fp16)[name = tensor<string, []>("op_1766_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_1766_cast_fp16, y = const_255_to_fp16)[name = tensor<string, []>("k_63_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_92_cast_fp16)[name = tensor<string, []>("op_1773_cast_fp16")];
            tensor<int32, [4]> var_1774 = const()[name = tensor<string, []>("op_1774"), 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, -3, -1])];
            tensor<int32, [4]> transpose_159_perm_0 = const()[name = tensor<string, []>("transpose_159_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1778_cast_fp16 = softmax(axis = var_1714, x = qk_31_cast_fp16)[name = tensor<string, []>("op_1778_cast_fp16")];
            tensor<bool, []> var_1780_transpose_x_0 = const()[name = tensor<string, []>("op_1780_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1780_transpose_y_0 = const()[name = tensor<string, []>("op_1780_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_259 = transpose(perm = var_1774, x = var_1773_cast_fp16)[name = tensor<string, []>("transpose_259")];
            tensor<fp16, [1, 20, 1500, 64]> var_1780_cast_fp16 = matmul(transpose_x = var_1780_transpose_x_0, transpose_y = var_1780_transpose_y_0, x = var_1778_cast_fp16, y = transpose_259)[name = tensor<string, []>("op_1780_cast_fp16")];
            tensor<int32, [4]> var_1781 = const()[name = tensor<string, []>("op_1781"), 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_1781, x = var_1780_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_1786_to_fp16 = const()[name = tensor<string, []>("op_1786_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(425615808)))];
            tensor<fp16, [1280]> var_1787_to_fp16 = const()[name = tensor<string, []>("op_1787_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428892672)))];
            tensor<fp16, [1, 1500, 1280]> linear_93_cast_fp16 = linear(bias = var_1787_to_fp16, weight = var_1786_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_1794_axes_0 = const()[name = tensor<string, []>("op_1794_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, []>(428895296)))];
            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, []>(428897920)))];
            tensor<fp16, [1, 1500, 1280]> var_1794_cast_fp16 = layer_norm(axes = var_1794_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1720_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16)[name = tensor<string, []>("op_1794_cast_fp16")];
            tensor<fp16, [4384, 1280]> var_1803_to_fp16 = const()[name = tensor<string, []>("op_1803_to_fp16"), val = tensor<fp16, [4384, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428900544)))];
            tensor<fp16, [4384]> var_1804_to_fp16 = const()[name = tensor<string, []>("op_1804_to_fp16"), val = tensor<fp16, [4384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440123648)))];
            tensor<fp16, [1, 1500, 4384]> linear_94_cast_fp16 = linear(bias = var_1804_to_fp16, weight = var_1803_to_fp16, x = var_1794_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, 4384]> 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, 4384]> var_1809_to_fp16 = const()[name = tensor<string, []>("op_1809_to_fp16"), val = tensor<fp16, [1280, 4384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(440132480)))];
            tensor<fp16, [1280]> var_1810_to_fp16 = const()[name = tensor<string, []>("op_1810_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(451355584)))];
            tensor<fp16, [1, 1500, 1280]> linear_95_cast_fp16 = linear(bias = var_1810_to_fp16, weight = var_1809_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_1820 = const()[name = tensor<string, []>("op_1820"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1836_axes_0 = const()[name = tensor<string, []>("op_1836_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, []>(451358208)))];
            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, []>(451360832)))];
            tensor<fp16, []> var_1826_to_fp16 = const()[name = tensor<string, []>("op_1826_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1836_cast_fp16 = layer_norm(axes = var_1836_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1826_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = tensor<string, []>("op_1836_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1847_to_fp16 = const()[name = tensor<string, []>("op_1847_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(451363456)))];
            tensor<fp16, [1280]> var_1848_to_fp16 = const()[name = tensor<string, []>("op_1848_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454640320)))];
            tensor<fp16, [1, 1500, 1280]> linear_96_cast_fp16 = linear(bias = var_1848_to_fp16, weight = var_1847_to_fp16, x = var_1836_cast_fp16)[name = tensor<string, []>("linear_96_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1851_to_fp16 = const()[name = tensor<string, []>("op_1851_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454642944)))];
            tensor<fp16, [1, 1500, 1280]> linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1851_to_fp16, x = var_1836_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1855_to_fp16 = const()[name = tensor<string, []>("op_1855_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457919808)))];
            tensor<fp16, [1280]> var_1856_to_fp16 = const()[name = tensor<string, []>("op_1856_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(461196672)))];
            tensor<fp16, [1, 1500, 1280]> linear_98_cast_fp16 = linear(bias = var_1856_to_fp16, weight = var_1855_to_fp16, x = var_1836_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")];
            tensor<int32, [4]> var_1864 = const()[name = tensor<string, []>("op_1864"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1865_cast_fp16 = reshape(shape = var_1864, x = linear_96_cast_fp16)[name = tensor<string, []>("op_1865_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_1865_cast_fp16, y = const_256_to_fp16)[name = tensor<string, []>("q_67_cast_fp16")];
            tensor<int32, [4]> var_1871 = const()[name = tensor<string, []>("op_1871"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1872_cast_fp16 = reshape(shape = var_1871, x = linear_97_cast_fp16)[name = tensor<string, []>("op_1872_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_1872_cast_fp16, y = const_257_to_fp16)[name = tensor<string, []>("k_67_cast_fp16")];
            tensor<int32, [4]> var_1878 = const()[name = tensor<string, []>("op_1878"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1879_cast_fp16 = reshape(shape = var_1878, x = linear_98_cast_fp16)[name = tensor<string, []>("op_1879_cast_fp16")];
            tensor<int32, [4]> var_1880 = const()[name = tensor<string, []>("op_1880"), 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, -3, -1])];
            tensor<int32, [4]> transpose_161_perm_0 = const()[name = tensor<string, []>("transpose_161_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1884_cast_fp16 = softmax(axis = var_1820, x = qk_33_cast_fp16)[name = tensor<string, []>("op_1884_cast_fp16")];
            tensor<bool, []> var_1886_transpose_x_0 = const()[name = tensor<string, []>("op_1886_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1886_transpose_y_0 = const()[name = tensor<string, []>("op_1886_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_255 = transpose(perm = var_1880, x = var_1879_cast_fp16)[name = tensor<string, []>("transpose_255")];
            tensor<fp16, [1, 20, 1500, 64]> var_1886_cast_fp16 = matmul(transpose_x = var_1886_transpose_x_0, transpose_y = var_1886_transpose_y_0, x = var_1884_cast_fp16, y = transpose_255)[name = tensor<string, []>("op_1886_cast_fp16")];
            tensor<int32, [4]> var_1887 = const()[name = tensor<string, []>("op_1887"), 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_1887, x = var_1886_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_1892_to_fp16 = const()[name = tensor<string, []>("op_1892_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(461199296)))];
            tensor<fp16, [1280]> var_1893_to_fp16 = const()[name = tensor<string, []>("op_1893_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464476160)))];
            tensor<fp16, [1, 1500, 1280]> linear_99_cast_fp16 = linear(bias = var_1893_to_fp16, weight = var_1892_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_1900_axes_0 = const()[name = tensor<string, []>("op_1900_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, []>(464478784)))];
            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, []>(464481408)))];
            tensor<fp16, [1, 1500, 1280]> var_1900_cast_fp16 = layer_norm(axes = var_1900_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1826_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = tensor<string, []>("op_1900_cast_fp16")];
            tensor<fp16, [4576, 1280]> var_1909_to_fp16 = const()[name = tensor<string, []>("op_1909_to_fp16"), val = tensor<fp16, [4576, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464484032)))];
            tensor<fp16, [4576]> var_1910_to_fp16 = const()[name = tensor<string, []>("op_1910_to_fp16"), val = tensor<fp16, [4576]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(476198656)))];
            tensor<fp16, [1, 1500, 4576]> linear_100_cast_fp16 = linear(bias = var_1910_to_fp16, weight = var_1909_to_fp16, x = var_1900_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, 4576]> 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, 4576]> var_1915_to_fp16 = const()[name = tensor<string, []>("op_1915_to_fp16"), val = tensor<fp16, [1280, 4576]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(476207872)))];
            tensor<fp16, [1280]> var_1916_to_fp16 = const()[name = tensor<string, []>("op_1916_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487922496)))];
            tensor<fp16, [1, 1500, 1280]> linear_101_cast_fp16 = linear(bias = var_1916_to_fp16, weight = var_1915_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_1926 = const()[name = tensor<string, []>("op_1926"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_1942_axes_0 = const()[name = tensor<string, []>("op_1942_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [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, []>(487925120)))];
            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, []>(487927744)))];
            tensor<fp16, []> var_1932_to_fp16 = const()[name = tensor<string, []>("op_1932_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_1942_cast_fp16 = layer_norm(axes = var_1942_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1932_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = tensor<string, []>("op_1942_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1953_to_fp16 = const()[name = tensor<string, []>("op_1953_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487930368)))];
            tensor<fp16, [1280]> var_1954_to_fp16 = const()[name = tensor<string, []>("op_1954_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491207232)))];
            tensor<fp16, [1, 1500, 1280]> linear_102_cast_fp16 = linear(bias = var_1954_to_fp16, weight = var_1953_to_fp16, x = var_1942_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1957_to_fp16 = const()[name = tensor<string, []>("op_1957_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(491209856)))];
            tensor<fp16, [1, 1500, 1280]> linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1957_to_fp16, x = var_1942_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_1961_to_fp16 = const()[name = tensor<string, []>("op_1961_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494486720)))];
            tensor<fp16, [1280]> var_1962_to_fp16 = const()[name = tensor<string, []>("op_1962_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(497763584)))];
            tensor<fp16, [1, 1500, 1280]> linear_104_cast_fp16 = linear(bias = var_1962_to_fp16, weight = var_1961_to_fp16, x = var_1942_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")];
            tensor<int32, [4]> var_1970 = const()[name = tensor<string, []>("op_1970"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1971_cast_fp16 = reshape(shape = var_1970, x = linear_102_cast_fp16)[name = tensor<string, []>("op_1971_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_1971_cast_fp16, y = const_258_to_fp16)[name = tensor<string, []>("q_71_cast_fp16")];
            tensor<int32, [4]> var_1977 = const()[name = tensor<string, []>("op_1977"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1978_cast_fp16 = reshape(shape = var_1977, x = linear_103_cast_fp16)[name = tensor<string, []>("op_1978_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_1978_cast_fp16, y = const_259_to_fp16)[name = tensor<string, []>("k_71_cast_fp16")];
            tensor<int32, [4]> var_1984 = const()[name = tensor<string, []>("op_1984"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_1985_cast_fp16 = reshape(shape = var_1984, x = linear_104_cast_fp16)[name = tensor<string, []>("op_1985_cast_fp16")];
            tensor<int32, [4]> var_1986 = const()[name = tensor<string, []>("op_1986"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> qk_35_transpose_x_0 = const()[name = tensor<string, []>("qk_35_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> qk_35_transpose_y_0 = const()[name = tensor<string, []>("qk_35_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_162_perm_0 = const()[name = tensor<string, []>("transpose_162_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_163_perm_0 = const()[name = tensor<string, []>("transpose_163_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_1990_cast_fp16 = softmax(axis = var_1926, x = qk_35_cast_fp16)[name = tensor<string, []>("op_1990_cast_fp16")];
            tensor<bool, []> var_1992_transpose_x_0 = const()[name = tensor<string, []>("op_1992_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1992_transpose_y_0 = const()[name = tensor<string, []>("op_1992_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_251 = transpose(perm = var_1986, x = var_1985_cast_fp16)[name = tensor<string, []>("transpose_251")];
            tensor<fp16, [1, 20, 1500, 64]> var_1992_cast_fp16 = matmul(transpose_x = var_1992_transpose_x_0, transpose_y = var_1992_transpose_y_0, x = var_1990_cast_fp16, y = transpose_251)[name = tensor<string, []>("op_1992_cast_fp16")];
            tensor<int32, [4]> var_1993 = const()[name = tensor<string, []>("op_1993"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_17 = const()[name = tensor<string, []>("concat_17"), val = tensor<int32, [3]>([1, 1500, 1280])];
            tensor<fp16, [1, 1500, 20, 64]> transpose_248 = transpose(perm = var_1993, x = var_1992_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_1998_to_fp16 = const()[name = tensor<string, []>("op_1998_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(497766208)))];
            tensor<fp16, [1280]> var_1999_to_fp16 = const()[name = tensor<string, []>("op_1999_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(501043072)))];
            tensor<fp16, [1, 1500, 1280]> linear_105_cast_fp16 = linear(bias = var_1999_to_fp16, weight = var_1998_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_2006_axes_0 = const()[name = tensor<string, []>("op_2006_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, []>(501045696)))];
            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, []>(501048320)))];
            tensor<fp16, [1, 1500, 1280]> var_2006_cast_fp16 = layer_norm(axes = var_2006_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1932_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = tensor<string, []>("op_2006_cast_fp16")];
            tensor<fp16, [4736, 1280]> var_2015_to_fp16 = const()[name = tensor<string, []>("op_2015_to_fp16"), val = tensor<fp16, [4736, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(501050944)))];
            tensor<fp16, [4736]> var_2016_to_fp16 = const()[name = tensor<string, []>("op_2016_to_fp16"), val = tensor<fp16, [4736]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513175168)))];
            tensor<fp16, [1, 1500, 4736]> linear_106_cast_fp16 = linear(bias = var_2016_to_fp16, weight = var_2015_to_fp16, x = var_2006_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, 4736]> 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, 4736]> var_2021_to_fp16 = const()[name = tensor<string, []>("op_2021_to_fp16"), val = tensor<fp16, [1280, 4736]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513184704)))];
            tensor<fp16, [1280]> var_2022_to_fp16 = const()[name = tensor<string, []>("op_2022_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(525308928)))];
            tensor<fp16, [1, 1500, 1280]> linear_107_cast_fp16 = linear(bias = var_2022_to_fp16, weight = var_2021_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_2032 = const()[name = tensor<string, []>("op_2032"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2048_axes_0 = const()[name = tensor<string, []>("op_2048_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, []>(525311552)))];
            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, []>(525314176)))];
            tensor<fp16, []> var_2038_to_fp16 = const()[name = tensor<string, []>("op_2038_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2048_cast_fp16 = layer_norm(axes = var_2048_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = tensor<string, []>("op_2048_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2059_to_fp16 = const()[name = tensor<string, []>("op_2059_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(525316800)))];
            tensor<fp16, [1280]> var_2060_to_fp16 = const()[name = tensor<string, []>("op_2060_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528593664)))];
            tensor<fp16, [1, 1500, 1280]> linear_108_cast_fp16 = linear(bias = var_2060_to_fp16, weight = var_2059_to_fp16, x = var_2048_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2063_to_fp16 = const()[name = tensor<string, []>("op_2063_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528596288)))];
            tensor<fp16, [1, 1500, 1280]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2063_to_fp16, x = var_2048_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2067_to_fp16 = const()[name = tensor<string, []>("op_2067_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(531873152)))];
            tensor<fp16, [1280]> var_2068_to_fp16 = const()[name = tensor<string, []>("op_2068_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(535150016)))];
            tensor<fp16, [1, 1500, 1280]> linear_110_cast_fp16 = linear(bias = var_2068_to_fp16, weight = var_2067_to_fp16, x = var_2048_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")];
            tensor<int32, [4]> var_2076 = const()[name = tensor<string, []>("op_2076"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2077_cast_fp16 = reshape(shape = var_2076, x = linear_108_cast_fp16)[name = tensor<string, []>("op_2077_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_2077_cast_fp16, y = const_260_to_fp16)[name = tensor<string, []>("q_75_cast_fp16")];
            tensor<int32, [4]> var_2083 = const()[name = tensor<string, []>("op_2083"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2084_cast_fp16 = reshape(shape = var_2083, x = linear_109_cast_fp16)[name = tensor<string, []>("op_2084_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_2084_cast_fp16, y = const_261_to_fp16)[name = tensor<string, []>("k_75_cast_fp16")];
            tensor<int32, [4]> var_2090 = const()[name = tensor<string, []>("op_2090"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2091_cast_fp16 = reshape(shape = var_2090, x = linear_110_cast_fp16)[name = tensor<string, []>("op_2091_cast_fp16")];
            tensor<int32, [4]> var_2092 = const()[name = tensor<string, []>("op_2092"), 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, -3, -1])];
            tensor<int32, [4]> transpose_165_perm_0 = const()[name = tensor<string, []>("transpose_165_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2096_cast_fp16 = softmax(axis = var_2032, x = qk_37_cast_fp16)[name = tensor<string, []>("op_2096_cast_fp16")];
            tensor<bool, []> var_2098_transpose_x_0 = const()[name = tensor<string, []>("op_2098_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2098_transpose_y_0 = const()[name = tensor<string, []>("op_2098_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_247 = transpose(perm = var_2092, x = var_2091_cast_fp16)[name = tensor<string, []>("transpose_247")];
            tensor<fp16, [1, 20, 1500, 64]> var_2098_cast_fp16 = matmul(transpose_x = var_2098_transpose_x_0, transpose_y = var_2098_transpose_y_0, x = var_2096_cast_fp16, y = transpose_247)[name = tensor<string, []>("op_2098_cast_fp16")];
            tensor<int32, [4]> var_2099 = const()[name = tensor<string, []>("op_2099"), 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_2099, x = var_2098_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_2104_to_fp16 = const()[name = tensor<string, []>("op_2104_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(535152640)))];
            tensor<fp16, [1280]> var_2105_to_fp16 = const()[name = tensor<string, []>("op_2105_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538429504)))];
            tensor<fp16, [1, 1500, 1280]> linear_111_cast_fp16 = linear(bias = var_2105_to_fp16, weight = var_2104_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_2112_axes_0 = const()[name = tensor<string, []>("op_2112_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, []>(538432128)))];
            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, []>(538434752)))];
            tensor<fp16, [1, 1500, 1280]> var_2112_cast_fp16 = layer_norm(axes = var_2112_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = tensor<string, []>("op_2112_cast_fp16")];
            tensor<fp16, [4672, 1280]> var_2121_to_fp16 = const()[name = tensor<string, []>("op_2121_to_fp16"), val = tensor<fp16, [4672, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538437376)))];
            tensor<fp16, [4672]> var_2122_to_fp16 = const()[name = tensor<string, []>("op_2122_to_fp16"), val = tensor<fp16, [4672]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550397760)))];
            tensor<fp16, [1, 1500, 4672]> linear_112_cast_fp16 = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = var_2112_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, 4672]> 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, 4672]> var_2127_to_fp16 = const()[name = tensor<string, []>("op_2127_to_fp16"), val = tensor<fp16, [1280, 4672]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550407168)))];
            tensor<fp16, [1280]> var_2128_to_fp16 = const()[name = tensor<string, []>("op_2128_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(562367552)))];
            tensor<fp16, [1, 1500, 1280]> linear_113_cast_fp16 = linear(bias = var_2128_to_fp16, weight = var_2127_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_2138 = const()[name = tensor<string, []>("op_2138"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2154_axes_0 = const()[name = tensor<string, []>("op_2154_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, []>(562370176)))];
            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, []>(562372800)))];
            tensor<fp16, []> var_2144_to_fp16 = const()[name = tensor<string, []>("op_2144_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2154_cast_fp16 = layer_norm(axes = var_2154_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2144_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = tensor<string, []>("op_2154_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2165_to_fp16 = const()[name = tensor<string, []>("op_2165_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(562375424)))];
            tensor<fp16, [1280]> var_2166_to_fp16 = const()[name = tensor<string, []>("op_2166_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565652288)))];
            tensor<fp16, [1, 1500, 1280]> linear_114_cast_fp16 = linear(bias = var_2166_to_fp16, weight = var_2165_to_fp16, x = var_2154_cast_fp16)[name = tensor<string, []>("linear_114_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2169_to_fp16 = const()[name = tensor<string, []>("op_2169_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565654912)))];
            tensor<fp16, [1, 1500, 1280]> linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2169_to_fp16, x = var_2154_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2173_to_fp16 = const()[name = tensor<string, []>("op_2173_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568931776)))];
            tensor<fp16, [1280]> var_2174_to_fp16 = const()[name = tensor<string, []>("op_2174_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572208640)))];
            tensor<fp16, [1, 1500, 1280]> linear_116_cast_fp16 = linear(bias = var_2174_to_fp16, weight = var_2173_to_fp16, x = var_2154_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")];
            tensor<int32, [4]> var_2182 = const()[name = tensor<string, []>("op_2182"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2183_cast_fp16 = reshape(shape = var_2182, x = linear_114_cast_fp16)[name = tensor<string, []>("op_2183_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_2183_cast_fp16, y = const_262_to_fp16)[name = tensor<string, []>("q_79_cast_fp16")];
            tensor<int32, [4]> var_2189 = const()[name = tensor<string, []>("op_2189"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2190_cast_fp16 = reshape(shape = var_2189, x = linear_115_cast_fp16)[name = tensor<string, []>("op_2190_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_2190_cast_fp16, y = const_263_to_fp16)[name = tensor<string, []>("k_79_cast_fp16")];
            tensor<int32, [4]> var_2196 = const()[name = tensor<string, []>("op_2196"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2197_cast_fp16 = reshape(shape = var_2196, x = linear_116_cast_fp16)[name = tensor<string, []>("op_2197_cast_fp16")];
            tensor<int32, [4]> var_2198 = const()[name = tensor<string, []>("op_2198"), 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, -3, -1])];
            tensor<int32, [4]> transpose_167_perm_0 = const()[name = tensor<string, []>("transpose_167_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2202_cast_fp16 = softmax(axis = var_2138, x = qk_39_cast_fp16)[name = tensor<string, []>("op_2202_cast_fp16")];
            tensor<bool, []> var_2204_transpose_x_0 = const()[name = tensor<string, []>("op_2204_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2204_transpose_y_0 = const()[name = tensor<string, []>("op_2204_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_243 = transpose(perm = var_2198, x = var_2197_cast_fp16)[name = tensor<string, []>("transpose_243")];
            tensor<fp16, [1, 20, 1500, 64]> var_2204_cast_fp16 = matmul(transpose_x = var_2204_transpose_x_0, transpose_y = var_2204_transpose_y_0, x = var_2202_cast_fp16, y = transpose_243)[name = tensor<string, []>("op_2204_cast_fp16")];
            tensor<int32, [4]> var_2205 = const()[name = tensor<string, []>("op_2205"), 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_2205, x = var_2204_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_2210_to_fp16 = const()[name = tensor<string, []>("op_2210_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572211264)))];
            tensor<fp16, [1280]> var_2211_to_fp16 = const()[name = tensor<string, []>("op_2211_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575488128)))];
            tensor<fp16, [1, 1500, 1280]> linear_117_cast_fp16 = linear(bias = var_2211_to_fp16, weight = var_2210_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_2218_axes_0 = const()[name = tensor<string, []>("op_2218_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, []>(575490752)))];
            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, []>(575493376)))];
            tensor<fp16, [1, 1500, 1280]> var_2218_cast_fp16 = layer_norm(axes = var_2218_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2144_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = tensor<string, []>("op_2218_cast_fp16")];
            tensor<fp16, [4800, 1280]> var_2227_to_fp16 = const()[name = tensor<string, []>("op_2227_to_fp16"), val = tensor<fp16, [4800, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575496000)))];
            tensor<fp16, [4800]> var_2228_to_fp16 = const()[name = tensor<string, []>("op_2228_to_fp16"), val = tensor<fp16, [4800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(587784064)))];
            tensor<fp16, [1, 1500, 4800]> linear_118_cast_fp16 = linear(bias = var_2228_to_fp16, weight = var_2227_to_fp16, x = var_2218_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, 4800]> 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, 4800]> var_2233_to_fp16 = const()[name = tensor<string, []>("op_2233_to_fp16"), val = tensor<fp16, [1280, 4800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(587793728)))];
            tensor<fp16, [1280]> var_2234_to_fp16 = const()[name = tensor<string, []>("op_2234_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(600081792)))];
            tensor<fp16, [1, 1500, 1280]> linear_119_cast_fp16 = linear(bias = var_2234_to_fp16, weight = var_2233_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_2244 = const()[name = tensor<string, []>("op_2244"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2260_axes_0 = const()[name = tensor<string, []>("op_2260_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, []>(600084416)))];
            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, []>(600087040)))];
            tensor<fp16, []> var_2250_to_fp16 = const()[name = tensor<string, []>("op_2250_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2260_cast_fp16 = layer_norm(axes = var_2260_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2250_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = tensor<string, []>("op_2260_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2271_to_fp16 = const()[name = tensor<string, []>("op_2271_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(600089664)))];
            tensor<fp16, [1280]> var_2272_to_fp16 = const()[name = tensor<string, []>("op_2272_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(603366528)))];
            tensor<fp16, [1, 1500, 1280]> linear_120_cast_fp16 = linear(bias = var_2272_to_fp16, weight = var_2271_to_fp16, x = var_2260_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2275_to_fp16 = const()[name = tensor<string, []>("op_2275_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(603369152)))];
            tensor<fp16, [1, 1500, 1280]> linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2275_to_fp16, x = var_2260_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2279_to_fp16 = const()[name = tensor<string, []>("op_2279_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606646016)))];
            tensor<fp16, [1280]> var_2280_to_fp16 = const()[name = tensor<string, []>("op_2280_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609922880)))];
            tensor<fp16, [1, 1500, 1280]> linear_122_cast_fp16 = linear(bias = var_2280_to_fp16, weight = var_2279_to_fp16, x = var_2260_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")];
            tensor<int32, [4]> var_2288 = const()[name = tensor<string, []>("op_2288"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2289_cast_fp16 = reshape(shape = var_2288, x = linear_120_cast_fp16)[name = tensor<string, []>("op_2289_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_2289_cast_fp16, y = const_264_to_fp16)[name = tensor<string, []>("q_83_cast_fp16")];
            tensor<int32, [4]> var_2295 = const()[name = tensor<string, []>("op_2295"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2296_cast_fp16 = reshape(shape = var_2295, x = linear_121_cast_fp16)[name = tensor<string, []>("op_2296_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_2296_cast_fp16, y = const_265_to_fp16)[name = tensor<string, []>("k_83_cast_fp16")];
            tensor<int32, [4]> var_2302 = const()[name = tensor<string, []>("op_2302"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2303_cast_fp16 = reshape(shape = var_2302, x = linear_122_cast_fp16)[name = tensor<string, []>("op_2303_cast_fp16")];
            tensor<int32, [4]> var_2304 = const()[name = tensor<string, []>("op_2304"), 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, -3, -1])];
            tensor<int32, [4]> transpose_169_perm_0 = const()[name = tensor<string, []>("transpose_169_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2308_cast_fp16 = softmax(axis = var_2244, x = qk_41_cast_fp16)[name = tensor<string, []>("op_2308_cast_fp16")];
            tensor<bool, []> var_2310_transpose_x_0 = const()[name = tensor<string, []>("op_2310_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2310_transpose_y_0 = const()[name = tensor<string, []>("op_2310_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_239 = transpose(perm = var_2304, x = var_2303_cast_fp16)[name = tensor<string, []>("transpose_239")];
            tensor<fp16, [1, 20, 1500, 64]> var_2310_cast_fp16 = matmul(transpose_x = var_2310_transpose_x_0, transpose_y = var_2310_transpose_y_0, x = var_2308_cast_fp16, y = transpose_239)[name = tensor<string, []>("op_2310_cast_fp16")];
            tensor<int32, [4]> var_2311 = const()[name = tensor<string, []>("op_2311"), 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_2311, x = var_2310_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_2316_to_fp16 = const()[name = tensor<string, []>("op_2316_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609925504)))];
            tensor<fp16, [1280]> var_2317_to_fp16 = const()[name = tensor<string, []>("op_2317_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(613202368)))];
            tensor<fp16, [1, 1500, 1280]> linear_123_cast_fp16 = linear(bias = var_2317_to_fp16, weight = var_2316_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_2324_axes_0 = const()[name = tensor<string, []>("op_2324_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, []>(613204992)))];
            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, []>(613207616)))];
            tensor<fp16, [1, 1500, 1280]> var_2324_cast_fp16 = layer_norm(axes = var_2324_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2250_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = tensor<string, []>("op_2324_cast_fp16")];
            tensor<fp16, [4864, 1280]> var_2333_to_fp16 = const()[name = tensor<string, []>("op_2333_to_fp16"), val = tensor<fp16, [4864, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(613210240)))];
            tensor<fp16, [4864]> var_2334_to_fp16 = const()[name = tensor<string, []>("op_2334_to_fp16"), val = tensor<fp16, [4864]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625662144)))];
            tensor<fp16, [1, 1500, 4864]> linear_124_cast_fp16 = linear(bias = var_2334_to_fp16, weight = var_2333_to_fp16, x = var_2324_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, 4864]> 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, 4864]> var_2339_to_fp16 = const()[name = tensor<string, []>("op_2339_to_fp16"), val = tensor<fp16, [1280, 4864]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625671936)))];
            tensor<fp16, [1280]> var_2340_to_fp16 = const()[name = tensor<string, []>("op_2340_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638123840)))];
            tensor<fp16, [1, 1500, 1280]> linear_125_cast_fp16 = linear(bias = var_2340_to_fp16, weight = var_2339_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_2350 = const()[name = tensor<string, []>("op_2350"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2366_axes_0 = const()[name = tensor<string, []>("op_2366_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, []>(638126464)))];
            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, []>(638129088)))];
            tensor<fp16, []> var_2356_to_fp16 = const()[name = tensor<string, []>("op_2356_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2366_cast_fp16 = layer_norm(axes = var_2366_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2356_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = tensor<string, []>("op_2366_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2377_to_fp16 = const()[name = tensor<string, []>("op_2377_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638131712)))];
            tensor<fp16, [1280]> var_2378_to_fp16 = const()[name = tensor<string, []>("op_2378_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641408576)))];
            tensor<fp16, [1, 1500, 1280]> linear_126_cast_fp16 = linear(bias = var_2378_to_fp16, weight = var_2377_to_fp16, x = var_2366_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2381_to_fp16 = const()[name = tensor<string, []>("op_2381_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641411200)))];
            tensor<fp16, [1, 1500, 1280]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2381_to_fp16, x = var_2366_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2385_to_fp16 = const()[name = tensor<string, []>("op_2385_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644688064)))];
            tensor<fp16, [1280]> var_2386_to_fp16 = const()[name = tensor<string, []>("op_2386_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(647964928)))];
            tensor<fp16, [1, 1500, 1280]> linear_128_cast_fp16 = linear(bias = var_2386_to_fp16, weight = var_2385_to_fp16, x = var_2366_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")];
            tensor<int32, [4]> var_2394 = const()[name = tensor<string, []>("op_2394"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2395_cast_fp16 = reshape(shape = var_2394, x = linear_126_cast_fp16)[name = tensor<string, []>("op_2395_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_2395_cast_fp16, y = const_266_to_fp16)[name = tensor<string, []>("q_87_cast_fp16")];
            tensor<int32, [4]> var_2401 = const()[name = tensor<string, []>("op_2401"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2402_cast_fp16 = reshape(shape = var_2401, x = linear_127_cast_fp16)[name = tensor<string, []>("op_2402_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_2402_cast_fp16, y = const_267_to_fp16)[name = tensor<string, []>("k_87_cast_fp16")];
            tensor<int32, [4]> var_2408 = const()[name = tensor<string, []>("op_2408"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2409_cast_fp16 = reshape(shape = var_2408, x = linear_128_cast_fp16)[name = tensor<string, []>("op_2409_cast_fp16")];
            tensor<int32, [4]> var_2410 = const()[name = tensor<string, []>("op_2410"), 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, -3, -1])];
            tensor<int32, [4]> transpose_171_perm_0 = const()[name = tensor<string, []>("transpose_171_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2414_cast_fp16 = softmax(axis = var_2350, x = qk_43_cast_fp16)[name = tensor<string, []>("op_2414_cast_fp16")];
            tensor<bool, []> var_2416_transpose_x_0 = const()[name = tensor<string, []>("op_2416_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2416_transpose_y_0 = const()[name = tensor<string, []>("op_2416_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_235 = transpose(perm = var_2410, x = var_2409_cast_fp16)[name = tensor<string, []>("transpose_235")];
            tensor<fp16, [1, 20, 1500, 64]> var_2416_cast_fp16 = matmul(transpose_x = var_2416_transpose_x_0, transpose_y = var_2416_transpose_y_0, x = var_2414_cast_fp16, y = transpose_235)[name = tensor<string, []>("op_2416_cast_fp16")];
            tensor<int32, [4]> var_2417 = const()[name = tensor<string, []>("op_2417"), 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_2417, x = var_2416_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_2422_to_fp16 = const()[name = tensor<string, []>("op_2422_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(647967552)))];
            tensor<fp16, [1280]> var_2423_to_fp16 = const()[name = tensor<string, []>("op_2423_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651244416)))];
            tensor<fp16, [1, 1500, 1280]> linear_129_cast_fp16 = linear(bias = var_2423_to_fp16, weight = var_2422_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_2430_axes_0 = const()[name = tensor<string, []>("op_2430_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, []>(651247040)))];
            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, []>(651249664)))];
            tensor<fp16, [1, 1500, 1280]> var_2430_cast_fp16 = layer_norm(axes = var_2430_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2356_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = tensor<string, []>("op_2430_cast_fp16")];
            tensor<fp16, [4928, 1280]> var_2439_to_fp16 = const()[name = tensor<string, []>("op_2439_to_fp16"), val = tensor<fp16, [4928, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651252288)))];
            tensor<fp16, [4928]> var_2440_to_fp16 = const()[name = tensor<string, []>("op_2440_to_fp16"), val = tensor<fp16, [4928]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(663868032)))];
            tensor<fp16, [1, 1500, 4928]> linear_130_cast_fp16 = linear(bias = var_2440_to_fp16, weight = var_2439_to_fp16, x = var_2430_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, 4928]> 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, 4928]> var_2445_to_fp16 = const()[name = tensor<string, []>("op_2445_to_fp16"), val = tensor<fp16, [1280, 4928]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(663877952)))];
            tensor<fp16, [1280]> var_2446_to_fp16 = const()[name = tensor<string, []>("op_2446_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(676493696)))];
            tensor<fp16, [1, 1500, 1280]> linear_131_cast_fp16 = linear(bias = var_2446_to_fp16, weight = var_2445_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_2456 = const()[name = tensor<string, []>("op_2456"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2472_axes_0 = const()[name = tensor<string, []>("op_2472_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, []>(676496320)))];
            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, []>(676498944)))];
            tensor<fp16, []> var_2462_to_fp16 = const()[name = tensor<string, []>("op_2462_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2472_cast_fp16 = layer_norm(axes = var_2472_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2462_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = tensor<string, []>("op_2472_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2483_to_fp16 = const()[name = tensor<string, []>("op_2483_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(676501568)))];
            tensor<fp16, [1280]> var_2484_to_fp16 = const()[name = tensor<string, []>("op_2484_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(679778432)))];
            tensor<fp16, [1, 1500, 1280]> linear_132_cast_fp16 = linear(bias = var_2484_to_fp16, weight = var_2483_to_fp16, x = var_2472_cast_fp16)[name = tensor<string, []>("linear_132_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2487_to_fp16 = const()[name = tensor<string, []>("op_2487_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(679781056)))];
            tensor<fp16, [1, 1500, 1280]> linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2487_to_fp16, x = var_2472_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2491_to_fp16 = const()[name = tensor<string, []>("op_2491_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683057920)))];
            tensor<fp16, [1280]> var_2492_to_fp16 = const()[name = tensor<string, []>("op_2492_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686334784)))];
            tensor<fp16, [1, 1500, 1280]> linear_134_cast_fp16 = linear(bias = var_2492_to_fp16, weight = var_2491_to_fp16, x = var_2472_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")];
            tensor<int32, [4]> var_2500 = const()[name = tensor<string, []>("op_2500"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2501_cast_fp16 = reshape(shape = var_2500, x = linear_132_cast_fp16)[name = tensor<string, []>("op_2501_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_2501_cast_fp16, y = const_268_to_fp16)[name = tensor<string, []>("q_91_cast_fp16")];
            tensor<int32, [4]> var_2507 = const()[name = tensor<string, []>("op_2507"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2508_cast_fp16 = reshape(shape = var_2507, x = linear_133_cast_fp16)[name = tensor<string, []>("op_2508_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_2508_cast_fp16, y = const_269_to_fp16)[name = tensor<string, []>("k_91_cast_fp16")];
            tensor<int32, [4]> var_2514 = const()[name = tensor<string, []>("op_2514"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2515_cast_fp16 = reshape(shape = var_2514, x = linear_134_cast_fp16)[name = tensor<string, []>("op_2515_cast_fp16")];
            tensor<int32, [4]> var_2516 = const()[name = tensor<string, []>("op_2516"), 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, -3, -1])];
            tensor<int32, [4]> transpose_173_perm_0 = const()[name = tensor<string, []>("transpose_173_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2520_cast_fp16 = softmax(axis = var_2456, x = qk_45_cast_fp16)[name = tensor<string, []>("op_2520_cast_fp16")];
            tensor<bool, []> var_2522_transpose_x_0 = const()[name = tensor<string, []>("op_2522_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2522_transpose_y_0 = const()[name = tensor<string, []>("op_2522_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_231 = transpose(perm = var_2516, x = var_2515_cast_fp16)[name = tensor<string, []>("transpose_231")];
            tensor<fp16, [1, 20, 1500, 64]> var_2522_cast_fp16 = matmul(transpose_x = var_2522_transpose_x_0, transpose_y = var_2522_transpose_y_0, x = var_2520_cast_fp16, y = transpose_231)[name = tensor<string, []>("op_2522_cast_fp16")];
            tensor<int32, [4]> var_2523 = const()[name = tensor<string, []>("op_2523"), 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_2523, x = var_2522_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_2528_to_fp16 = const()[name = tensor<string, []>("op_2528_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686337408)))];
            tensor<fp16, [1280]> var_2529_to_fp16 = const()[name = tensor<string, []>("op_2529_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(689614272)))];
            tensor<fp16, [1, 1500, 1280]> linear_135_cast_fp16 = linear(bias = var_2529_to_fp16, weight = var_2528_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_2536_axes_0 = const()[name = tensor<string, []>("op_2536_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, []>(689616896)))];
            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, []>(689619520)))];
            tensor<fp16, [1, 1500, 1280]> var_2536_cast_fp16 = layer_norm(axes = var_2536_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2462_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = tensor<string, []>("op_2536_cast_fp16")];
            tensor<fp16, [4992, 1280]> var_2545_to_fp16 = const()[name = tensor<string, []>("op_2545_to_fp16"), val = tensor<fp16, [4992, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(689622144)))];
            tensor<fp16, [4992]> var_2546_to_fp16 = const()[name = tensor<string, []>("op_2546_to_fp16"), val = tensor<fp16, [4992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(702401728)))];
            tensor<fp16, [1, 1500, 4992]> linear_136_cast_fp16 = linear(bias = var_2546_to_fp16, weight = var_2545_to_fp16, x = var_2536_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, 4992]> 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, 4992]> var_2551_to_fp16 = const()[name = tensor<string, []>("op_2551_to_fp16"), val = tensor<fp16, [1280, 4992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(702411776)))];
            tensor<fp16, [1280]> var_2552_to_fp16 = const()[name = tensor<string, []>("op_2552_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(715191360)))];
            tensor<fp16, [1, 1500, 1280]> linear_137_cast_fp16 = linear(bias = var_2552_to_fp16, weight = var_2551_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_2562 = const()[name = tensor<string, []>("op_2562"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2578_axes_0 = const()[name = tensor<string, []>("op_2578_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, []>(715193984)))];
            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, []>(715196608)))];
            tensor<fp16, []> var_2568_to_fp16 = const()[name = tensor<string, []>("op_2568_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2578_cast_fp16 = layer_norm(axes = var_2578_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2568_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = tensor<string, []>("op_2578_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2589_to_fp16 = const()[name = tensor<string, []>("op_2589_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(715199232)))];
            tensor<fp16, [1280]> var_2590_to_fp16 = const()[name = tensor<string, []>("op_2590_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(718476096)))];
            tensor<fp16, [1, 1500, 1280]> linear_138_cast_fp16 = linear(bias = var_2590_to_fp16, weight = var_2589_to_fp16, x = var_2578_cast_fp16)[name = tensor<string, []>("linear_138_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2593_to_fp16 = const()[name = tensor<string, []>("op_2593_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(718478720)))];
            tensor<fp16, [1, 1500, 1280]> linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2593_to_fp16, x = var_2578_cast_fp16)[name = tensor<string, []>("linear_139_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2597_to_fp16 = const()[name = tensor<string, []>("op_2597_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721755584)))];
            tensor<fp16, [1280]> var_2598_to_fp16 = const()[name = tensor<string, []>("op_2598_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(725032448)))];
            tensor<fp16, [1, 1500, 1280]> linear_140_cast_fp16 = linear(bias = var_2598_to_fp16, weight = var_2597_to_fp16, x = var_2578_cast_fp16)[name = tensor<string, []>("linear_140_cast_fp16")];
            tensor<int32, [4]> var_2606 = const()[name = tensor<string, []>("op_2606"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2607_cast_fp16 = reshape(shape = var_2606, x = linear_138_cast_fp16)[name = tensor<string, []>("op_2607_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_2607_cast_fp16, y = const_270_to_fp16)[name = tensor<string, []>("q_95_cast_fp16")];
            tensor<int32, [4]> var_2613 = const()[name = tensor<string, []>("op_2613"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2614_cast_fp16 = reshape(shape = var_2613, x = linear_139_cast_fp16)[name = tensor<string, []>("op_2614_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_2614_cast_fp16, y = const_271_to_fp16)[name = tensor<string, []>("k_95_cast_fp16")];
            tensor<int32, [4]> var_2620 = const()[name = tensor<string, []>("op_2620"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2621_cast_fp16 = reshape(shape = var_2620, x = linear_140_cast_fp16)[name = tensor<string, []>("op_2621_cast_fp16")];
            tensor<int32, [4]> var_2622 = const()[name = tensor<string, []>("op_2622"), 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, -3, -1])];
            tensor<int32, [4]> transpose_175_perm_0 = const()[name = tensor<string, []>("transpose_175_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2626_cast_fp16 = softmax(axis = var_2562, x = qk_47_cast_fp16)[name = tensor<string, []>("op_2626_cast_fp16")];
            tensor<bool, []> var_2628_transpose_x_0 = const()[name = tensor<string, []>("op_2628_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2628_transpose_y_0 = const()[name = tensor<string, []>("op_2628_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_227 = transpose(perm = var_2622, x = var_2621_cast_fp16)[name = tensor<string, []>("transpose_227")];
            tensor<fp16, [1, 20, 1500, 64]> var_2628_cast_fp16 = matmul(transpose_x = var_2628_transpose_x_0, transpose_y = var_2628_transpose_y_0, x = var_2626_cast_fp16, y = transpose_227)[name = tensor<string, []>("op_2628_cast_fp16")];
            tensor<int32, [4]> var_2629 = const()[name = tensor<string, []>("op_2629"), 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_2629, x = var_2628_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_2634_to_fp16 = const()[name = tensor<string, []>("op_2634_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(725035072)))];
            tensor<fp16, [1280]> var_2635_to_fp16 = const()[name = tensor<string, []>("op_2635_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728311936)))];
            tensor<fp16, [1, 1500, 1280]> linear_141_cast_fp16 = linear(bias = var_2635_to_fp16, weight = var_2634_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_2642_axes_0 = const()[name = tensor<string, []>("op_2642_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, []>(728314560)))];
            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, []>(728317184)))];
            tensor<fp16, [1, 1500, 1280]> var_2642_cast_fp16 = layer_norm(axes = var_2642_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2568_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = tensor<string, []>("op_2642_cast_fp16")];
            tensor<fp16, [5024, 1280]> var_2651_to_fp16 = const()[name = tensor<string, []>("op_2651_to_fp16"), val = tensor<fp16, [5024, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728319808)))];
            tensor<fp16, [5024]> var_2652_to_fp16 = const()[name = tensor<string, []>("op_2652_to_fp16"), val = tensor<fp16, [5024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(741181312)))];
            tensor<fp16, [1, 1500, 5024]> linear_142_cast_fp16 = linear(bias = var_2652_to_fp16, weight = var_2651_to_fp16, x = var_2642_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, 5024]> 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, 5024]> var_2657_to_fp16 = const()[name = tensor<string, []>("op_2657_to_fp16"), val = tensor<fp16, [1280, 5024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(741191424)))];
            tensor<fp16, [1280]> var_2658_to_fp16 = const()[name = tensor<string, []>("op_2658_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754052928)))];
            tensor<fp16, [1, 1500, 1280]> linear_143_cast_fp16 = linear(bias = var_2658_to_fp16, weight = var_2657_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_2668 = const()[name = tensor<string, []>("op_2668"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2684_axes_0 = const()[name = tensor<string, []>("op_2684_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, []>(754055552)))];
            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, []>(754058176)))];
            tensor<fp16, []> var_2674_to_fp16 = const()[name = tensor<string, []>("op_2674_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2684_cast_fp16 = layer_norm(axes = var_2684_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2674_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast_fp16)[name = tensor<string, []>("op_2684_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2695_to_fp16 = const()[name = tensor<string, []>("op_2695_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(754060800)))];
            tensor<fp16, [1280]> var_2696_to_fp16 = const()[name = tensor<string, []>("op_2696_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757337664)))];
            tensor<fp16, [1, 1500, 1280]> linear_144_cast_fp16 = linear(bias = var_2696_to_fp16, weight = var_2695_to_fp16, x = var_2684_cast_fp16)[name = tensor<string, []>("linear_144_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2699_to_fp16 = const()[name = tensor<string, []>("op_2699_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757340288)))];
            tensor<fp16, [1, 1500, 1280]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2699_to_fp16, x = var_2684_cast_fp16)[name = tensor<string, []>("linear_145_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2703_to_fp16 = const()[name = tensor<string, []>("op_2703_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(760617152)))];
            tensor<fp16, [1280]> var_2704_to_fp16 = const()[name = tensor<string, []>("op_2704_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(763894016)))];
            tensor<fp16, [1, 1500, 1280]> linear_146_cast_fp16 = linear(bias = var_2704_to_fp16, weight = var_2703_to_fp16, x = var_2684_cast_fp16)[name = tensor<string, []>("linear_146_cast_fp16")];
            tensor<int32, [4]> var_2712 = const()[name = tensor<string, []>("op_2712"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2713_cast_fp16 = reshape(shape = var_2712, x = linear_144_cast_fp16)[name = tensor<string, []>("op_2713_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_2713_cast_fp16, y = const_272_to_fp16)[name = tensor<string, []>("q_99_cast_fp16")];
            tensor<int32, [4]> var_2719 = const()[name = tensor<string, []>("op_2719"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2720_cast_fp16 = reshape(shape = var_2719, x = linear_145_cast_fp16)[name = tensor<string, []>("op_2720_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_2720_cast_fp16, y = const_273_to_fp16)[name = tensor<string, []>("k_99_cast_fp16")];
            tensor<int32, [4]> var_2726 = const()[name = tensor<string, []>("op_2726"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2727_cast_fp16 = reshape(shape = var_2726, x = linear_146_cast_fp16)[name = tensor<string, []>("op_2727_cast_fp16")];
            tensor<int32, [4]> var_2728 = const()[name = tensor<string, []>("op_2728"), 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, -3, -1])];
            tensor<int32, [4]> transpose_177_perm_0 = const()[name = tensor<string, []>("transpose_177_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2732_cast_fp16 = softmax(axis = var_2668, x = qk_49_cast_fp16)[name = tensor<string, []>("op_2732_cast_fp16")];
            tensor<bool, []> var_2734_transpose_x_0 = const()[name = tensor<string, []>("op_2734_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2734_transpose_y_0 = const()[name = tensor<string, []>("op_2734_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_223 = transpose(perm = var_2728, x = var_2727_cast_fp16)[name = tensor<string, []>("transpose_223")];
            tensor<fp16, [1, 20, 1500, 64]> var_2734_cast_fp16 = matmul(transpose_x = var_2734_transpose_x_0, transpose_y = var_2734_transpose_y_0, x = var_2732_cast_fp16, y = transpose_223)[name = tensor<string, []>("op_2734_cast_fp16")];
            tensor<int32, [4]> var_2735 = const()[name = tensor<string, []>("op_2735"), 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_2735, x = var_2734_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_2740_to_fp16 = const()[name = tensor<string, []>("op_2740_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(763896640)))];
            tensor<fp16, [1280]> var_2741_to_fp16 = const()[name = tensor<string, []>("op_2741_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(767173504)))];
            tensor<fp16, [1, 1500, 1280]> linear_147_cast_fp16 = linear(bias = var_2741_to_fp16, weight = var_2740_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_2748_axes_0 = const()[name = tensor<string, []>("op_2748_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, []>(767176128)))];
            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, []>(767178752)))];
            tensor<fp16, [1, 1500, 1280]> var_2748_cast_fp16 = layer_norm(axes = var_2748_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2674_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast_fp16)[name = tensor<string, []>("op_2748_cast_fp16")];
            tensor<fp16, [5024, 1280]> var_2757_to_fp16 = const()[name = tensor<string, []>("op_2757_to_fp16"), val = tensor<fp16, [5024, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(767181376)))];
            tensor<fp16, [5024]> var_2758_to_fp16 = const()[name = tensor<string, []>("op_2758_to_fp16"), val = tensor<fp16, [5024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780042880)))];
            tensor<fp16, [1, 1500, 5024]> linear_148_cast_fp16 = linear(bias = var_2758_to_fp16, weight = var_2757_to_fp16, x = var_2748_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, 5024]> 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, 5024]> var_2763_to_fp16 = const()[name = tensor<string, []>("op_2763_to_fp16"), val = tensor<fp16, [1280, 5024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780052992)))];
            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, []>(792914496)))];
            tensor<fp16, [1, 1500, 1280]> linear_149_cast_fp16 = linear(bias = var_2764_to_fp16, weight = var_2763_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_2774 = const()[name = tensor<string, []>("op_2774"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2790_axes_0 = const()[name = tensor<string, []>("op_2790_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, []>(792917120)))];
            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, []>(792919744)))];
            tensor<fp16, []> var_2780_to_fp16 = const()[name = tensor<string, []>("op_2780_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2790_cast_fp16 = layer_norm(axes = var_2790_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2780_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast_fp16)[name = tensor<string, []>("op_2790_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2801_to_fp16 = const()[name = tensor<string, []>("op_2801_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792922368)))];
            tensor<fp16, [1280]> var_2802_to_fp16 = const()[name = tensor<string, []>("op_2802_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(796199232)))];
            tensor<fp16, [1, 1500, 1280]> linear_150_cast_fp16 = linear(bias = var_2802_to_fp16, weight = var_2801_to_fp16, x = var_2790_cast_fp16)[name = tensor<string, []>("linear_150_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2805_to_fp16 = const()[name = tensor<string, []>("op_2805_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(796201856)))];
            tensor<fp16, [1, 1500, 1280]> linear_151_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2805_to_fp16, x = var_2790_cast_fp16)[name = tensor<string, []>("linear_151_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2809_to_fp16 = const()[name = tensor<string, []>("op_2809_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(799478720)))];
            tensor<fp16, [1280]> var_2810_to_fp16 = const()[name = tensor<string, []>("op_2810_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(802755584)))];
            tensor<fp16, [1, 1500, 1280]> linear_152_cast_fp16 = linear(bias = var_2810_to_fp16, weight = var_2809_to_fp16, x = var_2790_cast_fp16)[name = tensor<string, []>("linear_152_cast_fp16")];
            tensor<int32, [4]> var_2818 = const()[name = tensor<string, []>("op_2818"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2819_cast_fp16 = reshape(shape = var_2818, x = linear_150_cast_fp16)[name = tensor<string, []>("op_2819_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_2819_cast_fp16, y = const_274_to_fp16)[name = tensor<string, []>("q_103_cast_fp16")];
            tensor<int32, [4]> var_2825 = const()[name = tensor<string, []>("op_2825"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2826_cast_fp16 = reshape(shape = var_2825, x = linear_151_cast_fp16)[name = tensor<string, []>("op_2826_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_2826_cast_fp16, y = const_275_to_fp16)[name = tensor<string, []>("k_103_cast_fp16")];
            tensor<int32, [4]> var_2832 = const()[name = tensor<string, []>("op_2832"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2833_cast_fp16 = reshape(shape = var_2832, x = linear_152_cast_fp16)[name = tensor<string, []>("op_2833_cast_fp16")];
            tensor<int32, [4]> var_2834 = const()[name = tensor<string, []>("op_2834"), 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, -3, -1])];
            tensor<int32, [4]> transpose_179_perm_0 = const()[name = tensor<string, []>("transpose_179_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2838_cast_fp16 = softmax(axis = var_2774, x = qk_51_cast_fp16)[name = tensor<string, []>("op_2838_cast_fp16")];
            tensor<bool, []> var_2840_transpose_x_0 = const()[name = tensor<string, []>("op_2840_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2840_transpose_y_0 = const()[name = tensor<string, []>("op_2840_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_219 = transpose(perm = var_2834, x = var_2833_cast_fp16)[name = tensor<string, []>("transpose_219")];
            tensor<fp16, [1, 20, 1500, 64]> var_2840_cast_fp16 = matmul(transpose_x = var_2840_transpose_x_0, transpose_y = var_2840_transpose_y_0, x = var_2838_cast_fp16, y = transpose_219)[name = tensor<string, []>("op_2840_cast_fp16")];
            tensor<int32, [4]> var_2841 = const()[name = tensor<string, []>("op_2841"), 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_2841, x = var_2840_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_2846_to_fp16 = const()[name = tensor<string, []>("op_2846_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(802758208)))];
            tensor<fp16, [1280]> var_2847_to_fp16 = const()[name = tensor<string, []>("op_2847_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(806035072)))];
            tensor<fp16, [1, 1500, 1280]> linear_153_cast_fp16 = linear(bias = var_2847_to_fp16, weight = var_2846_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_2854_axes_0 = const()[name = tensor<string, []>("op_2854_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, []>(806037696)))];
            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, []>(806040320)))];
            tensor<fp16, [1, 1500, 1280]> var_2854_cast_fp16 = layer_norm(axes = var_2854_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2780_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast_fp16)[name = tensor<string, []>("op_2854_cast_fp16")];
            tensor<fp16, [4992, 1280]> var_2863_to_fp16 = const()[name = tensor<string, []>("op_2863_to_fp16"), val = tensor<fp16, [4992, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(806042944)))];
            tensor<fp16, [4992]> var_2864_to_fp16 = const()[name = tensor<string, []>("op_2864_to_fp16"), val = tensor<fp16, [4992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(818822528)))];
            tensor<fp16, [1, 1500, 4992]> linear_154_cast_fp16 = linear(bias = var_2864_to_fp16, weight = var_2863_to_fp16, x = var_2854_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, 4992]> 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, 4992]> var_2869_to_fp16 = const()[name = tensor<string, []>("op_2869_to_fp16"), val = tensor<fp16, [1280, 4992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(818832576)))];
            tensor<fp16, [1280]> var_2870_to_fp16 = const()[name = tensor<string, []>("op_2870_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(831612160)))];
            tensor<fp16, [1, 1500, 1280]> linear_155_cast_fp16 = linear(bias = var_2870_to_fp16, weight = var_2869_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_2880 = const()[name = tensor<string, []>("op_2880"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_2896_axes_0 = const()[name = tensor<string, []>("op_2896_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, []>(831614784)))];
            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, []>(831617408)))];
            tensor<fp16, []> var_2886_to_fp16 = const()[name = tensor<string, []>("op_2886_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_2896_cast_fp16 = layer_norm(axes = var_2896_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2886_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast_fp16)[name = tensor<string, []>("op_2896_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2907_to_fp16 = const()[name = tensor<string, []>("op_2907_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(831620032)))];
            tensor<fp16, [1280]> var_2908_to_fp16 = const()[name = tensor<string, []>("op_2908_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(834896896)))];
            tensor<fp16, [1, 1500, 1280]> linear_156_cast_fp16 = linear(bias = var_2908_to_fp16, weight = var_2907_to_fp16, x = var_2896_cast_fp16)[name = tensor<string, []>("linear_156_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2911_to_fp16 = const()[name = tensor<string, []>("op_2911_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(834899520)))];
            tensor<fp16, [1, 1500, 1280]> linear_157_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2911_to_fp16, x = var_2896_cast_fp16)[name = tensor<string, []>("linear_157_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_2915_to_fp16 = const()[name = tensor<string, []>("op_2915_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838176384)))];
            tensor<fp16, [1280]> var_2916_to_fp16 = const()[name = tensor<string, []>("op_2916_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841453248)))];
            tensor<fp16, [1, 1500, 1280]> linear_158_cast_fp16 = linear(bias = var_2916_to_fp16, weight = var_2915_to_fp16, x = var_2896_cast_fp16)[name = tensor<string, []>("linear_158_cast_fp16")];
            tensor<int32, [4]> var_2924 = const()[name = tensor<string, []>("op_2924"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2925_cast_fp16 = reshape(shape = var_2924, x = linear_156_cast_fp16)[name = tensor<string, []>("op_2925_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_2925_cast_fp16, y = const_276_to_fp16)[name = tensor<string, []>("q_107_cast_fp16")];
            tensor<int32, [4]> var_2931 = const()[name = tensor<string, []>("op_2931"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2932_cast_fp16 = reshape(shape = var_2931, x = linear_157_cast_fp16)[name = tensor<string, []>("op_2932_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_2932_cast_fp16, y = const_277_to_fp16)[name = tensor<string, []>("k_107_cast_fp16")];
            tensor<int32, [4]> var_2938 = const()[name = tensor<string, []>("op_2938"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_2939_cast_fp16 = reshape(shape = var_2938, x = linear_158_cast_fp16)[name = tensor<string, []>("op_2939_cast_fp16")];
            tensor<int32, [4]> var_2940 = const()[name = tensor<string, []>("op_2940"), 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, -3, -1])];
            tensor<int32, [4]> transpose_181_perm_0 = const()[name = tensor<string, []>("transpose_181_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_2944_cast_fp16 = softmax(axis = var_2880, x = qk_53_cast_fp16)[name = tensor<string, []>("op_2944_cast_fp16")];
            tensor<bool, []> var_2946_transpose_x_0 = const()[name = tensor<string, []>("op_2946_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2946_transpose_y_0 = const()[name = tensor<string, []>("op_2946_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_215 = transpose(perm = var_2940, x = var_2939_cast_fp16)[name = tensor<string, []>("transpose_215")];
            tensor<fp16, [1, 20, 1500, 64]> var_2946_cast_fp16 = matmul(transpose_x = var_2946_transpose_x_0, transpose_y = var_2946_transpose_y_0, x = var_2944_cast_fp16, y = transpose_215)[name = tensor<string, []>("op_2946_cast_fp16")];
            tensor<int32, [4]> var_2947 = const()[name = tensor<string, []>("op_2947"), 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_2947, x = var_2946_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_2952_to_fp16 = const()[name = tensor<string, []>("op_2952_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841455872)))];
            tensor<fp16, [1280]> var_2953_to_fp16 = const()[name = tensor<string, []>("op_2953_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844732736)))];
            tensor<fp16, [1, 1500, 1280]> linear_159_cast_fp16 = linear(bias = var_2953_to_fp16, weight = var_2952_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_2960_axes_0 = const()[name = tensor<string, []>("op_2960_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, []>(844735360)))];
            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, []>(844737984)))];
            tensor<fp16, [1, 1500, 1280]> var_2960_cast_fp16 = layer_norm(axes = var_2960_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2886_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast_fp16)[name = tensor<string, []>("op_2960_cast_fp16")];
            tensor<fp16, [4992, 1280]> var_2969_to_fp16 = const()[name = tensor<string, []>("op_2969_to_fp16"), val = tensor<fp16, [4992, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844740608)))];
            tensor<fp16, [4992]> var_2970_to_fp16 = const()[name = tensor<string, []>("op_2970_to_fp16"), val = tensor<fp16, [4992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(857520192)))];
            tensor<fp16, [1, 1500, 4992]> linear_160_cast_fp16 = linear(bias = var_2970_to_fp16, weight = var_2969_to_fp16, x = var_2960_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, 4992]> 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, 4992]> var_2975_to_fp16 = const()[name = tensor<string, []>("op_2975_to_fp16"), val = tensor<fp16, [1280, 4992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(857530240)))];
            tensor<fp16, [1280]> var_2976_to_fp16 = const()[name = tensor<string, []>("op_2976_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(870309824)))];
            tensor<fp16, [1, 1500, 1280]> linear_161_cast_fp16 = linear(bias = var_2976_to_fp16, weight = var_2975_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_2986 = const()[name = tensor<string, []>("op_2986"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3002_axes_0 = const()[name = tensor<string, []>("op_3002_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, []>(870312448)))];
            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, []>(870315072)))];
            tensor<fp16, []> var_2992_to_fp16 = const()[name = tensor<string, []>("op_2992_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3002_cast_fp16 = layer_norm(axes = var_3002_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_2992_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast_fp16)[name = tensor<string, []>("op_3002_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3013_to_fp16 = const()[name = tensor<string, []>("op_3013_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(870317696)))];
            tensor<fp16, [1280]> var_3014_to_fp16 = const()[name = tensor<string, []>("op_3014_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(873594560)))];
            tensor<fp16, [1, 1500, 1280]> linear_162_cast_fp16 = linear(bias = var_3014_to_fp16, weight = var_3013_to_fp16, x = var_3002_cast_fp16)[name = tensor<string, []>("linear_162_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3017_to_fp16 = const()[name = tensor<string, []>("op_3017_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(873597184)))];
            tensor<fp16, [1, 1500, 1280]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3017_to_fp16, x = var_3002_cast_fp16)[name = tensor<string, []>("linear_163_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3021_to_fp16 = const()[name = tensor<string, []>("op_3021_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(876874048)))];
            tensor<fp16, [1280]> var_3022_to_fp16 = const()[name = tensor<string, []>("op_3022_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880150912)))];
            tensor<fp16, [1, 1500, 1280]> linear_164_cast_fp16 = linear(bias = var_3022_to_fp16, weight = var_3021_to_fp16, x = var_3002_cast_fp16)[name = tensor<string, []>("linear_164_cast_fp16")];
            tensor<int32, [4]> var_3030 = const()[name = tensor<string, []>("op_3030"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3031_cast_fp16 = reshape(shape = var_3030, x = linear_162_cast_fp16)[name = tensor<string, []>("op_3031_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_3031_cast_fp16, y = const_278_to_fp16)[name = tensor<string, []>("q_111_cast_fp16")];
            tensor<int32, [4]> var_3037 = const()[name = tensor<string, []>("op_3037"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3038_cast_fp16 = reshape(shape = var_3037, x = linear_163_cast_fp16)[name = tensor<string, []>("op_3038_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_3038_cast_fp16, y = const_279_to_fp16)[name = tensor<string, []>("k_111_cast_fp16")];
            tensor<int32, [4]> var_3044 = const()[name = tensor<string, []>("op_3044"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3045_cast_fp16 = reshape(shape = var_3044, x = linear_164_cast_fp16)[name = tensor<string, []>("op_3045_cast_fp16")];
            tensor<int32, [4]> var_3046 = const()[name = tensor<string, []>("op_3046"), 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, -3, -1])];
            tensor<int32, [4]> transpose_183_perm_0 = const()[name = tensor<string, []>("transpose_183_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_3050_cast_fp16 = softmax(axis = var_2986, x = qk_55_cast_fp16)[name = tensor<string, []>("op_3050_cast_fp16")];
            tensor<bool, []> var_3052_transpose_x_0 = const()[name = tensor<string, []>("op_3052_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3052_transpose_y_0 = const()[name = tensor<string, []>("op_3052_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_211 = transpose(perm = var_3046, x = var_3045_cast_fp16)[name = tensor<string, []>("transpose_211")];
            tensor<fp16, [1, 20, 1500, 64]> var_3052_cast_fp16 = matmul(transpose_x = var_3052_transpose_x_0, transpose_y = var_3052_transpose_y_0, x = var_3050_cast_fp16, y = transpose_211)[name = tensor<string, []>("op_3052_cast_fp16")];
            tensor<int32, [4]> var_3053 = const()[name = tensor<string, []>("op_3053"), 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_3053, x = var_3052_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_3058_to_fp16 = const()[name = tensor<string, []>("op_3058_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880153536)))];
            tensor<fp16, [1280]> var_3059_to_fp16 = const()[name = tensor<string, []>("op_3059_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883430400)))];
            tensor<fp16, [1, 1500, 1280]> linear_165_cast_fp16 = linear(bias = var_3059_to_fp16, weight = var_3058_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_3066_axes_0 = const()[name = tensor<string, []>("op_3066_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, []>(883433024)))];
            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, []>(883435648)))];
            tensor<fp16, [1, 1500, 1280]> var_3066_cast_fp16 = layer_norm(axes = var_3066_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_2992_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast_fp16)[name = tensor<string, []>("op_3066_cast_fp16")];
            tensor<fp16, [5024, 1280]> var_3075_to_fp16 = const()[name = tensor<string, []>("op_3075_to_fp16"), val = tensor<fp16, [5024, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883438272)))];
            tensor<fp16, [5024]> var_3076_to_fp16 = const()[name = tensor<string, []>("op_3076_to_fp16"), val = tensor<fp16, [5024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896299776)))];
            tensor<fp16, [1, 1500, 5024]> linear_166_cast_fp16 = linear(bias = var_3076_to_fp16, weight = var_3075_to_fp16, x = var_3066_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, 5024]> 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, 5024]> var_3081_to_fp16 = const()[name = tensor<string, []>("op_3081_to_fp16"), val = tensor<fp16, [1280, 5024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896309888)))];
            tensor<fp16, [1280]> var_3082_to_fp16 = const()[name = tensor<string, []>("op_3082_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909171392)))];
            tensor<fp16, [1, 1500, 1280]> linear_167_cast_fp16 = linear(bias = var_3082_to_fp16, weight = var_3081_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_3092 = const()[name = tensor<string, []>("op_3092"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3108_axes_0 = const()[name = tensor<string, []>("op_3108_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, []>(909174016)))];
            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, []>(909176640)))];
            tensor<fp16, []> var_3098_to_fp16 = const()[name = tensor<string, []>("op_3098_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3108_cast_fp16 = layer_norm(axes = var_3108_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3098_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast_fp16)[name = tensor<string, []>("op_3108_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3119_to_fp16 = const()[name = tensor<string, []>("op_3119_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(909179264)))];
            tensor<fp16, [1280]> var_3120_to_fp16 = const()[name = tensor<string, []>("op_3120_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(912456128)))];
            tensor<fp16, [1, 1500, 1280]> linear_168_cast_fp16 = linear(bias = var_3120_to_fp16, weight = var_3119_to_fp16, x = var_3108_cast_fp16)[name = tensor<string, []>("linear_168_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3123_to_fp16 = const()[name = tensor<string, []>("op_3123_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(912458752)))];
            tensor<fp16, [1, 1500, 1280]> linear_169_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3123_to_fp16, x = var_3108_cast_fp16)[name = tensor<string, []>("linear_169_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3127_to_fp16 = const()[name = tensor<string, []>("op_3127_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(915735616)))];
            tensor<fp16, [1280]> var_3128_to_fp16 = const()[name = tensor<string, []>("op_3128_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919012480)))];
            tensor<fp16, [1, 1500, 1280]> linear_170_cast_fp16 = linear(bias = var_3128_to_fp16, weight = var_3127_to_fp16, x = var_3108_cast_fp16)[name = tensor<string, []>("linear_170_cast_fp16")];
            tensor<int32, [4]> var_3136 = const()[name = tensor<string, []>("op_3136"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3137_cast_fp16 = reshape(shape = var_3136, x = linear_168_cast_fp16)[name = tensor<string, []>("op_3137_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_3137_cast_fp16, y = const_280_to_fp16)[name = tensor<string, []>("q_115_cast_fp16")];
            tensor<int32, [4]> var_3143 = const()[name = tensor<string, []>("op_3143"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3144_cast_fp16 = reshape(shape = var_3143, x = linear_169_cast_fp16)[name = tensor<string, []>("op_3144_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_3144_cast_fp16, y = const_281_to_fp16)[name = tensor<string, []>("k_115_cast_fp16")];
            tensor<int32, [4]> var_3150 = const()[name = tensor<string, []>("op_3150"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3151_cast_fp16 = reshape(shape = var_3150, x = linear_170_cast_fp16)[name = tensor<string, []>("op_3151_cast_fp16")];
            tensor<int32, [4]> var_3152 = const()[name = tensor<string, []>("op_3152"), 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, -3, -1])];
            tensor<int32, [4]> transpose_185_perm_0 = const()[name = tensor<string, []>("transpose_185_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_3156_cast_fp16 = softmax(axis = var_3092, x = qk_57_cast_fp16)[name = tensor<string, []>("op_3156_cast_fp16")];
            tensor<bool, []> var_3158_transpose_x_0 = const()[name = tensor<string, []>("op_3158_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3158_transpose_y_0 = const()[name = tensor<string, []>("op_3158_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_207 = transpose(perm = var_3152, x = var_3151_cast_fp16)[name = tensor<string, []>("transpose_207")];
            tensor<fp16, [1, 20, 1500, 64]> var_3158_cast_fp16 = matmul(transpose_x = var_3158_transpose_x_0, transpose_y = var_3158_transpose_y_0, x = var_3156_cast_fp16, y = transpose_207)[name = tensor<string, []>("op_3158_cast_fp16")];
            tensor<int32, [4]> var_3159 = const()[name = tensor<string, []>("op_3159"), 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_3159, x = var_3158_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_3164_to_fp16 = const()[name = tensor<string, []>("op_3164_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919015104)))];
            tensor<fp16, [1280]> var_3165_to_fp16 = const()[name = tensor<string, []>("op_3165_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(922291968)))];
            tensor<fp16, [1, 1500, 1280]> linear_171_cast_fp16 = linear(bias = var_3165_to_fp16, weight = var_3164_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_3172_axes_0 = const()[name = tensor<string, []>("op_3172_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, []>(922294592)))];
            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, []>(922297216)))];
            tensor<fp16, [1, 1500, 1280]> var_3172_cast_fp16 = layer_norm(axes = var_3172_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3098_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast_fp16)[name = tensor<string, []>("op_3172_cast_fp16")];
            tensor<fp16, [5056, 1280]> var_3181_to_fp16 = const()[name = tensor<string, []>("op_3181_to_fp16"), val = tensor<fp16, [5056, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(922299840)))];
            tensor<fp16, [5056]> var_3182_to_fp16 = const()[name = tensor<string, []>("op_3182_to_fp16"), val = tensor<fp16, [5056]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935243264)))];
            tensor<fp16, [1, 1500, 5056]> linear_172_cast_fp16 = linear(bias = var_3182_to_fp16, weight = var_3181_to_fp16, x = var_3172_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, 5056]> 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, 5056]> var_3187_to_fp16 = const()[name = tensor<string, []>("op_3187_to_fp16"), val = tensor<fp16, [1280, 5056]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935253440)))];
            tensor<fp16, [1280]> var_3188_to_fp16 = const()[name = tensor<string, []>("op_3188_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(948196864)))];
            tensor<fp16, [1, 1500, 1280]> linear_173_cast_fp16 = linear(bias = var_3188_to_fp16, weight = var_3187_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_3198 = const()[name = tensor<string, []>("op_3198"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3214_axes_0 = const()[name = tensor<string, []>("op_3214_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, []>(948199488)))];
            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, []>(948202112)))];
            tensor<fp16, []> var_3204_to_fp16 = const()[name = tensor<string, []>("op_3204_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3214_cast_fp16 = layer_norm(axes = var_3214_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3204_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast_fp16)[name = tensor<string, []>("op_3214_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3225_to_fp16 = const()[name = tensor<string, []>("op_3225_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(948204736)))];
            tensor<fp16, [1280]> var_3226_to_fp16 = const()[name = tensor<string, []>("op_3226_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951481600)))];
            tensor<fp16, [1, 1500, 1280]> linear_174_cast_fp16 = linear(bias = var_3226_to_fp16, weight = var_3225_to_fp16, x = var_3214_cast_fp16)[name = tensor<string, []>("linear_174_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3229_to_fp16 = const()[name = tensor<string, []>("op_3229_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(951484224)))];
            tensor<fp16, [1, 1500, 1280]> linear_175_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3229_to_fp16, x = var_3214_cast_fp16)[name = tensor<string, []>("linear_175_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3233_to_fp16 = const()[name = tensor<string, []>("op_3233_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954761088)))];
            tensor<fp16, [1280]> var_3234_to_fp16 = const()[name = tensor<string, []>("op_3234_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958037952)))];
            tensor<fp16, [1, 1500, 1280]> linear_176_cast_fp16 = linear(bias = var_3234_to_fp16, weight = var_3233_to_fp16, x = var_3214_cast_fp16)[name = tensor<string, []>("linear_176_cast_fp16")];
            tensor<int32, [4]> var_3242 = const()[name = tensor<string, []>("op_3242"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3243_cast_fp16 = reshape(shape = var_3242, x = linear_174_cast_fp16)[name = tensor<string, []>("op_3243_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_3243_cast_fp16, y = const_282_to_fp16)[name = tensor<string, []>("q_119_cast_fp16")];
            tensor<int32, [4]> var_3249 = const()[name = tensor<string, []>("op_3249"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3250_cast_fp16 = reshape(shape = var_3249, x = linear_175_cast_fp16)[name = tensor<string, []>("op_3250_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_3250_cast_fp16, y = const_283_to_fp16)[name = tensor<string, []>("k_119_cast_fp16")];
            tensor<int32, [4]> var_3256 = const()[name = tensor<string, []>("op_3256"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3257_cast_fp16 = reshape(shape = var_3256, x = linear_176_cast_fp16)[name = tensor<string, []>("op_3257_cast_fp16")];
            tensor<int32, [4]> var_3258 = const()[name = tensor<string, []>("op_3258"), 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, -3, -1])];
            tensor<int32, [4]> transpose_187_perm_0 = const()[name = tensor<string, []>("transpose_187_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_3262_cast_fp16 = softmax(axis = var_3198, x = qk_59_cast_fp16)[name = tensor<string, []>("op_3262_cast_fp16")];
            tensor<bool, []> var_3264_transpose_x_0 = const()[name = tensor<string, []>("op_3264_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3264_transpose_y_0 = const()[name = tensor<string, []>("op_3264_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_203 = transpose(perm = var_3258, x = var_3257_cast_fp16)[name = tensor<string, []>("transpose_203")];
            tensor<fp16, [1, 20, 1500, 64]> var_3264_cast_fp16 = matmul(transpose_x = var_3264_transpose_x_0, transpose_y = var_3264_transpose_y_0, x = var_3262_cast_fp16, y = transpose_203)[name = tensor<string, []>("op_3264_cast_fp16")];
            tensor<int32, [4]> var_3265 = const()[name = tensor<string, []>("op_3265"), 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_3265, x = var_3264_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_3270_to_fp16 = const()[name = tensor<string, []>("op_3270_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958040576)))];
            tensor<fp16, [1280]> var_3271_to_fp16 = const()[name = tensor<string, []>("op_3271_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(961317440)))];
            tensor<fp16, [1, 1500, 1280]> linear_177_cast_fp16 = linear(bias = var_3271_to_fp16, weight = var_3270_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_3278_axes_0 = const()[name = tensor<string, []>("op_3278_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, []>(961320064)))];
            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, []>(961322688)))];
            tensor<fp16, [1, 1500, 1280]> var_3278_cast_fp16 = layer_norm(axes = var_3278_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3204_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast_fp16)[name = tensor<string, []>("op_3278_cast_fp16")];
            tensor<fp16, [5024, 1280]> var_3287_to_fp16 = const()[name = tensor<string, []>("op_3287_to_fp16"), val = tensor<fp16, [5024, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(961325312)))];
            tensor<fp16, [5024]> var_3288_to_fp16 = const()[name = tensor<string, []>("op_3288_to_fp16"), val = tensor<fp16, [5024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(974186816)))];
            tensor<fp16, [1, 1500, 5024]> linear_178_cast_fp16 = linear(bias = var_3288_to_fp16, weight = var_3287_to_fp16, x = var_3278_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, 5024]> 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, 5024]> var_3293_to_fp16 = const()[name = tensor<string, []>("op_3293_to_fp16"), val = tensor<fp16, [1280, 5024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(974196928)))];
            tensor<fp16, [1280]> var_3294_to_fp16 = const()[name = tensor<string, []>("op_3294_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(987058432)))];
            tensor<fp16, [1, 1500, 1280]> linear_179_cast_fp16 = linear(bias = var_3294_to_fp16, weight = var_3293_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_3304 = const()[name = tensor<string, []>("op_3304"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3320_axes_0 = const()[name = tensor<string, []>("op_3320_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, []>(987061056)))];
            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, []>(987063680)))];
            tensor<fp16, []> var_3310_to_fp16 = const()[name = tensor<string, []>("op_3310_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3320_cast_fp16 = layer_norm(axes = var_3320_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3310_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast_fp16)[name = tensor<string, []>("op_3320_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3331_to_fp16 = const()[name = tensor<string, []>("op_3331_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(987066304)))];
            tensor<fp16, [1280]> var_3332_to_fp16 = const()[name = tensor<string, []>("op_3332_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(990343168)))];
            tensor<fp16, [1, 1500, 1280]> linear_180_cast_fp16 = linear(bias = var_3332_to_fp16, weight = var_3331_to_fp16, x = var_3320_cast_fp16)[name = tensor<string, []>("linear_180_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3335_to_fp16 = const()[name = tensor<string, []>("op_3335_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(990345792)))];
            tensor<fp16, [1, 1500, 1280]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3335_to_fp16, x = var_3320_cast_fp16)[name = tensor<string, []>("linear_181_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3339_to_fp16 = const()[name = tensor<string, []>("op_3339_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(993622656)))];
            tensor<fp16, [1280]> var_3340_to_fp16 = const()[name = tensor<string, []>("op_3340_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(996899520)))];
            tensor<fp16, [1, 1500, 1280]> linear_182_cast_fp16 = linear(bias = var_3340_to_fp16, weight = var_3339_to_fp16, x = var_3320_cast_fp16)[name = tensor<string, []>("linear_182_cast_fp16")];
            tensor<int32, [4]> var_3348 = const()[name = tensor<string, []>("op_3348"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3349_cast_fp16 = reshape(shape = var_3348, x = linear_180_cast_fp16)[name = tensor<string, []>("op_3349_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_3349_cast_fp16, y = const_284_to_fp16)[name = tensor<string, []>("q_123_cast_fp16")];
            tensor<int32, [4]> var_3355 = const()[name = tensor<string, []>("op_3355"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3356_cast_fp16 = reshape(shape = var_3355, x = linear_181_cast_fp16)[name = tensor<string, []>("op_3356_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_3356_cast_fp16, y = const_285_to_fp16)[name = tensor<string, []>("k_123_cast_fp16")];
            tensor<int32, [4]> var_3362 = const()[name = tensor<string, []>("op_3362"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3363_cast_fp16 = reshape(shape = var_3362, x = linear_182_cast_fp16)[name = tensor<string, []>("op_3363_cast_fp16")];
            tensor<int32, [4]> var_3364 = const()[name = tensor<string, []>("op_3364"), 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, -3, -1])];
            tensor<int32, [4]> transpose_189_perm_0 = const()[name = tensor<string, []>("transpose_189_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_3368_cast_fp16 = softmax(axis = var_3304, x = qk_61_cast_fp16)[name = tensor<string, []>("op_3368_cast_fp16")];
            tensor<bool, []> var_3370_transpose_x_0 = const()[name = tensor<string, []>("op_3370_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3370_transpose_y_0 = const()[name = tensor<string, []>("op_3370_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_199 = transpose(perm = var_3364, x = var_3363_cast_fp16)[name = tensor<string, []>("transpose_199")];
            tensor<fp16, [1, 20, 1500, 64]> var_3370_cast_fp16 = matmul(transpose_x = var_3370_transpose_x_0, transpose_y = var_3370_transpose_y_0, x = var_3368_cast_fp16, y = transpose_199)[name = tensor<string, []>("op_3370_cast_fp16")];
            tensor<int32, [4]> var_3371 = const()[name = tensor<string, []>("op_3371"), 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_3371, x = var_3370_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_3376_to_fp16 = const()[name = tensor<string, []>("op_3376_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(996902144)))];
            tensor<fp16, [1280]> var_3377_to_fp16 = const()[name = tensor<string, []>("op_3377_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1000179008)))];
            tensor<fp16, [1, 1500, 1280]> linear_183_cast_fp16 = linear(bias = var_3377_to_fp16, weight = var_3376_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_3384_axes_0 = const()[name = tensor<string, []>("op_3384_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, []>(1000181632)))];
            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, []>(1000184256)))];
            tensor<fp16, [1, 1500, 1280]> var_3384_cast_fp16 = layer_norm(axes = var_3384_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3310_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast_fp16)[name = tensor<string, []>("op_3384_cast_fp16")];
            tensor<fp16, [4992, 1280]> var_3393_to_fp16 = const()[name = tensor<string, []>("op_3393_to_fp16"), val = tensor<fp16, [4992, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1000186880)))];
            tensor<fp16, [4992]> var_3394_to_fp16 = const()[name = tensor<string, []>("op_3394_to_fp16"), val = tensor<fp16, [4992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1012966464)))];
            tensor<fp16, [1, 1500, 4992]> linear_184_cast_fp16 = linear(bias = var_3394_to_fp16, weight = var_3393_to_fp16, x = var_3384_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, 4992]> 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, 4992]> var_3399_to_fp16 = const()[name = tensor<string, []>("op_3399_to_fp16"), val = tensor<fp16, [1280, 4992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1012976512)))];
            tensor<fp16, [1280]> var_3400_to_fp16 = const()[name = tensor<string, []>("op_3400_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1025756096)))];
            tensor<fp16, [1, 1500, 1280]> linear_185_cast_fp16 = linear(bias = var_3400_to_fp16, weight = var_3399_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_3410 = const()[name = tensor<string, []>("op_3410"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_3426_axes_0 = const()[name = tensor<string, []>("op_3426_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, []>(1025758720)))];
            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, []>(1025761344)))];
            tensor<fp16, []> var_3416_to_fp16 = const()[name = tensor<string, []>("op_3416_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> var_3426_cast_fp16 = layer_norm(axes = var_3426_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3416_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast_fp16)[name = tensor<string, []>("op_3426_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3437_to_fp16 = const()[name = tensor<string, []>("op_3437_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1025763968)))];
            tensor<fp16, [1280]> var_3438_to_fp16 = const()[name = tensor<string, []>("op_3438_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1029040832)))];
            tensor<fp16, [1, 1500, 1280]> linear_186_cast_fp16 = linear(bias = var_3438_to_fp16, weight = var_3437_to_fp16, x = var_3426_cast_fp16)[name = tensor<string, []>("linear_186_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3441_to_fp16 = const()[name = tensor<string, []>("op_3441_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1029043456)))];
            tensor<fp16, [1, 1500, 1280]> linear_187_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3441_to_fp16, x = var_3426_cast_fp16)[name = tensor<string, []>("linear_187_cast_fp16")];
            tensor<fp16, [1280, 1280]> var_3445_to_fp16 = const()[name = tensor<string, []>("op_3445_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1032320320)))];
            tensor<fp16, [1280]> var_3446_to_fp16 = const()[name = tensor<string, []>("op_3446_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1035597184)))];
            tensor<fp16, [1, 1500, 1280]> linear_188_cast_fp16 = linear(bias = var_3446_to_fp16, weight = var_3445_to_fp16, x = var_3426_cast_fp16)[name = tensor<string, []>("linear_188_cast_fp16")];
            tensor<int32, [4]> var_3454 = const()[name = tensor<string, []>("op_3454"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3455_cast_fp16 = reshape(shape = var_3454, x = linear_186_cast_fp16)[name = tensor<string, []>("op_3455_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_3455_cast_fp16, y = const_286_to_fp16)[name = tensor<string, []>("q_cast_fp16")];
            tensor<int32, [4]> var_3461 = const()[name = tensor<string, []>("op_3461"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3462_cast_fp16 = reshape(shape = var_3461, x = linear_187_cast_fp16)[name = tensor<string, []>("op_3462_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_3462_cast_fp16, y = const_287_to_fp16)[name = tensor<string, []>("k_cast_fp16")];
            tensor<int32, [4]> var_3468 = const()[name = tensor<string, []>("op_3468"), val = tensor<int32, [4]>([1, 1500, 20, -1])];
            tensor<fp16, [1, 1500, 20, 64]> var_3469_cast_fp16 = reshape(shape = var_3468, x = linear_188_cast_fp16)[name = tensor<string, []>("op_3469_cast_fp16")];
            tensor<int32, [4]> var_3470 = const()[name = tensor<string, []>("op_3470"), 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, -3, -1])];
            tensor<int32, [4]> transpose_191_perm_0 = const()[name = tensor<string, []>("transpose_191_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            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_3474_cast_fp16 = softmax(axis = var_3410, x = qk_cast_fp16)[name = tensor<string, []>("op_3474_cast_fp16")];
            tensor<bool, []> var_3476_transpose_x_0 = const()[name = tensor<string, []>("op_3476_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3476_transpose_y_0 = const()[name = tensor<string, []>("op_3476_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 20, 1500, 64]> transpose_195 = transpose(perm = var_3470, x = var_3469_cast_fp16)[name = tensor<string, []>("transpose_195")];
            tensor<fp16, [1, 20, 1500, 64]> var_3476_cast_fp16 = matmul(transpose_x = var_3476_transpose_x_0, transpose_y = var_3476_transpose_y_0, x = var_3474_cast_fp16, y = transpose_195)[name = tensor<string, []>("op_3476_cast_fp16")];
            tensor<int32, [4]> var_3477 = const()[name = tensor<string, []>("op_3477"), 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_3477, x = var_3476_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_3482_to_fp16 = const()[name = tensor<string, []>("op_3482_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1035599808)))];
            tensor<fp16, [1280]> var_3483_to_fp16 = const()[name = tensor<string, []>("op_3483_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1038876672)))];
            tensor<fp16, [1, 1500, 1280]> linear_189_cast_fp16 = linear(bias = var_3483_to_fp16, weight = var_3482_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_3490_axes_0 = const()[name = tensor<string, []>("op_3490_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, []>(1038879296)))];
            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, []>(1038881920)))];
            tensor<fp16, [1, 1500, 1280]> var_3490_cast_fp16 = layer_norm(axes = var_3490_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3416_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast_fp16)[name = tensor<string, []>("op_3490_cast_fp16")];
            tensor<fp16, [4864, 1280]> var_3499_to_fp16 = const()[name = tensor<string, []>("op_3499_to_fp16"), val = tensor<fp16, [4864, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1038884544)))];
            tensor<fp16, [4864]> var_3500_to_fp16 = const()[name = tensor<string, []>("op_3500_to_fp16"), val = tensor<fp16, [4864]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1051336448)))];
            tensor<fp16, [1, 1500, 4864]> linear_190_cast_fp16 = linear(bias = var_3500_to_fp16, weight = var_3499_to_fp16, x = var_3490_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, 4864]> 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, 4864]> var_3505_to_fp16 = const()[name = tensor<string, []>("op_3505_to_fp16"), val = tensor<fp16, [1280, 4864]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1051346240)))];
            tensor<fp16, [1280]> var_3506_to_fp16 = const()[name = tensor<string, []>("op_3506_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1063798144)))];
            tensor<fp16, [1, 1500, 1280]> linear_191_cast_fp16 = linear(bias = var_3506_to_fp16, weight = var_3505_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_3519_axes_0 = const()[name = tensor<string, []>("op_3519_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, []>(1063800768)))];
            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, []>(1063803392)))];
            tensor<fp16, []> var_3510_to_fp16 = const()[name = tensor<string, []>("op_3510_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1280]> output = layer_norm(axes = var_3519_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3510_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = tensor<string, []>("op_3519_cast_fp16")];
        } -> (output);
}