program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.9.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
    func main<ios17>(tensor<int32, [1, 256]> attention_mask, tensor<int32, [1, 256]> input_ids) {
            tensor<int32, []> var_16 = const()[name = tensor<string, []>("op_16"), val = tensor<int32, []>(0)];
            tensor<int32, []> var_18 = const()[name = tensor<string, []>("op_18"), val = tensor<int32, []>(-1)];
            tensor<int32, []> var_19 = const()[name = tensor<string, []>("op_19"), val = tensor<int32, []>(1)];
            tensor<int32, []> var_55_batch_dims_0 = const()[name = tensor<string, []>("op_55_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> var_55_validate_indices_0 = const()[name = tensor<string, []>("op_55_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [256206, 1024]> encoder_embed_tokens_weight_to_fp16 = const()[name = tensor<string, []>("encoder_embed_tokens_weight_to_fp16"), val = tensor<fp16, [256206, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<int32, []> greater_equal_0_y_0 = const()[name = tensor<string, []>("greater_equal_0_y_0"), val = tensor<int32, []>(0)];
            tensor<bool, [1, 256]> greater_equal_0 = greater_equal(x = input_ids, y = greater_equal_0_y_0)[name = tensor<string, []>("greater_equal_0")];
            tensor<int32, []> slice_by_index_0 = const()[name = tensor<string, []>("slice_by_index_0"), val = tensor<int32, []>(256206)];
            tensor<int32, [1, 256]> add_0 = add(x = input_ids, y = slice_by_index_0)[name = tensor<string, []>("add_0")];
            tensor<int32, [1, 256]> select_0 = select(a = input_ids, b = add_0, cond = greater_equal_0)[name = tensor<string, []>("select_0")];
            tensor<int32, []> var_55_cast_fp16_axis_0 = const()[name = tensor<string, []>("op_55_cast_fp16_axis_0"), val = tensor<int32, []>(0)];
            tensor<fp16, [1, 256, 1024]> var_55_cast_fp16 = gather(axis = var_55_cast_fp16_axis_0, batch_dims = var_55_batch_dims_0, indices = select_0, validate_indices = var_55_validate_indices_0, x = encoder_embed_tokens_weight_to_fp16)[name = tensor<string, []>("op_55_cast_fp16")];
            tensor<fp16, []> var_56_to_fp16 = const()[name = tensor<string, []>("op_56_to_fp16"), val = tensor<fp16, []>(0x1p+5)];
            tensor<fp16, [1, 256, 1024]> inputs_embeds_1_cast_fp16 = mul(x = var_55_cast_fp16, y = var_56_to_fp16)[name = tensor<string, []>("inputs_embeds_1_cast_fp16")];
            tensor<bool, [1, 256]> var_61 = not_equal(x = input_ids, y = var_19)[name = tensor<string, []>("op_61")];
            tensor<string, []> mask_1_dtype_0 = const()[name = tensor<string, []>("mask_1_dtype_0"), val = tensor<string, []>("int32")];
            tensor<bool, []> var_63_exclusive_0 = const()[name = tensor<string, []>("op_63_exclusive_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_63_reverse_0 = const()[name = tensor<string, []>("op_63_reverse_0"), val = tensor<bool, []>(false)];
            tensor<int32, [1, 256]> mask_1 = cast(dtype = mask_1_dtype_0, x = var_61)[name = tensor<string, []>("cast_56")];
            tensor<int32, [1, 256]> var_63 = cumsum(axis = var_19, exclusive = var_63_exclusive_0, reverse = var_63_reverse_0, x = mask_1)[name = tensor<string, []>("op_63")];
            tensor<int32, [1, 256]> incremental_indices_1 = mul(x = var_63, y = mask_1)[name = tensor<string, []>("incremental_indices_1")];
            tensor<int32, []> var_69 = const()[name = tensor<string, []>("op_69"), val = tensor<int32, []>(1)];
            tensor<int32, [1, 256]> var_70 = add(x = incremental_indices_1, y = var_69)[name = tensor<string, []>("op_70")];
            tensor<int32, [1]> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<int32, [1]>([-1])];
            tensor<int32, [256]> var_73 = reshape(shape = var_72, x = var_70)[name = tensor<string, []>("op_73")];
            tensor<int32, []> var_74_batch_dims_0 = const()[name = tensor<string, []>("op_74_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> var_74_validate_indices_0 = const()[name = tensor<string, []>("op_74_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1026, 1024]> encoder_embed_positions_weights_to_fp16 = const()[name = tensor<string, []>("encoder_embed_positions_weights_to_fp16"), val = tensor<fp16, [1026, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524710016)))];
            tensor<string, []> var_73_to_uint16_dtype_0 = const()[name = tensor<string, []>("op_73_to_uint16_dtype_0"), val = tensor<string, []>("uint16")];
            tensor<uint16, [256]> var_73_to_uint16 = cast(dtype = var_73_to_uint16_dtype_0, x = var_73)[name = tensor<string, []>("cast_55")];
            tensor<fp16, [256, 1024]> var_74_cast_fp16_cast_uint16 = gather(axis = var_16, batch_dims = var_74_batch_dims_0, indices = var_73_to_uint16, validate_indices = var_74_validate_indices_0, x = encoder_embed_positions_weights_to_fp16)[name = tensor<string, []>("op_74_cast_fp16_cast_uint16")];
            tensor<int32, [3]> var_76 = const()[name = tensor<string, []>("op_76"), val = tensor<int32, [3]>([1, 256, 1024])];
            tensor<fp16, [1, 256, 1024]> var_77_cast_fp16 = reshape(shape = var_76, x = var_74_cast_fp16_cast_uint16)[name = tensor<string, []>("op_77_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input0_7_cast_fp16 = add(x = inputs_embeds_1_cast_fp16, y = var_77_cast_fp16)[name = tensor<string, []>("input0_7_cast_fp16")];
            tensor<int32, [1]> var_85_axes_0 = const()[name = tensor<string, []>("op_85_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1, 1, 256]> var_85 = expand_dims(axes = var_85_axes_0, x = attention_mask)[name = tensor<string, []>("op_85")];
            tensor<int32, [1]> var_86_axes_0 = const()[name = tensor<string, []>("op_86_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1, 1, 1, 256]> var_86 = expand_dims(axes = var_86_axes_0, x = var_85)[name = tensor<string, []>("op_86")];
            tensor<int32, [4]> var_89_reps_0 = const()[name = tensor<string, []>("op_89_reps_0"), val = tensor<int32, [4]>([1, 1, 256, 1])];
            tensor<int32, [1, 1, 256, 256]> var_89 = tile(reps = var_89_reps_0, x = var_86)[name = tensor<string, []>("op_89")];
            tensor<fp16, []> const_6_to_fp16 = const()[name = tensor<string, []>("const_6_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<string, []> expanded_mask_1_to_fp16_dtype_0 = const()[name = tensor<string, []>("expanded_mask_1_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1, 1, 256, 256]> var_89_to_fp16 = cast(dtype = expanded_mask_1_to_fp16_dtype_0, x = var_89)[name = tensor<string, []>("cast_54")];
            tensor<fp16, [1, 1, 256, 256]> inverted_mask_1_cast_fp16 = sub(x = const_6_to_fp16, y = var_89_to_fp16)[name = tensor<string, []>("inverted_mask_1_cast_fp16")];
            tensor<string, []> var_94_dtype_0 = const()[name = tensor<string, []>("op_94_dtype_0"), val = tensor<string, []>("bool")];
            tensor<fp16, []> var_10_to_fp16 = const()[name = tensor<string, []>("op_10_to_fp16"), val = tensor<fp16, []>(-inf)];
            tensor<bool, [1, 1, 256, 256]> inverted_mask_1_cast_fp16_to_bool = cast(dtype = var_94_dtype_0, x = inverted_mask_1_cast_fp16)[name = tensor<string, []>("cast_53")];
            tensor<fp16, [1, 1, 256, 256]> attention_mask0_1_cast_fp16 = select(a = var_10_to_fp16, b = inverted_mask_1_cast_fp16, cond = inverted_mask_1_cast_fp16_to_bool)[name = tensor<string, []>("attention_mask0_1_cast_fp16")];
            tensor<int32, [1]> hidden_states_3_axes_0 = const()[name = tensor<string, []>("hidden_states_3_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_0_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526811328)))];
            tensor<fp16, [1024]> encoder_layers_0_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526813440)))];
            tensor<fp16, []> var_6_to_fp16 = const()[name = tensor<string, []>("op_6_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 256, 1024]> hidden_states_3_cast_fp16 = layer_norm(axes = hidden_states_3_axes_0, beta = encoder_layers_0_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_0_self_attn_layer_norm_weight_to_fp16, x = input0_7_cast_fp16)[name = tensor<string, []>("hidden_states_3_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526815552)))];
            tensor<fp16, [1024]> encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528912768)))];
            tensor<fp16, [1, 256, 1024]> linear_0_cast_fp16 = linear(bias = encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_q_proj_weight_to_fp16, x = hidden_states_3_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<int32, [4]> var_114 = const()[name = tensor<string, []>("op_114"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_115_cast_fp16 = reshape(shape = var_114, x = linear_0_cast_fp16)[name = tensor<string, []>("op_115_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528914880)))];
            tensor<fp16, [1024]> encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(531012096)))];
            tensor<fp16, [1, 256, 1024]> linear_1_cast_fp16 = linear(bias = encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_k_proj_weight_to_fp16, x = hidden_states_3_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(531014208)))];
            tensor<fp16, [1024]> encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533111424)))];
            tensor<fp16, [1, 256, 1024]> linear_2_cast_fp16 = linear(bias = encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_v_proj_weight_to_fp16, x = hidden_states_3_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> var_123 = const()[name = tensor<string, []>("op_123"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_124_cast_fp16 = reshape(shape = var_123, x = linear_1_cast_fp16)[name = tensor<string, []>("op_124_cast_fp16")];
            tensor<int32, [4]> var_126 = const()[name = tensor<string, []>("op_126"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_127_cast_fp16 = reshape(shape = var_126, x = linear_2_cast_fp16)[name = tensor<string, []>("op_127_cast_fp16")];
            tensor<int32, [4]> value_2_perm_0 = const()[name = tensor<string, []>("value_2_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_130_transpose_x_0 = const()[name = tensor<string, []>("op_130_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_130_transpose_y_0 = const()[name = tensor<string, []>("op_130_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_36_perm_0 = const()[name = tensor<string, []>("transpose_36_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_37_perm_0 = const()[name = tensor<string, []>("transpose_37_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_37 = transpose(perm = transpose_37_perm_0, x = var_124_cast_fp16)[name = tensor<string, []>("transpose_105")];
            tensor<fp16, [1, 16, 256, 64]> transpose_36 = transpose(perm = transpose_36_perm_0, x = var_115_cast_fp16)[name = tensor<string, []>("transpose_106")];
            tensor<fp16, [1, 16, 256, 256]> var_130_cast_fp16 = matmul(transpose_x = var_130_transpose_x_0, transpose_y = var_130_transpose_y_0, x = transpose_36, y = transpose_37)[name = tensor<string, []>("op_130_cast_fp16")];
            tensor<fp16, []> var_131_to_fp16 = const()[name = tensor<string, []>("op_131_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_2_cast_fp16 = mul(x = var_130_cast_fp16, y = var_131_to_fp16)[name = tensor<string, []>("attn_weights_2_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_5_cast_fp16 = add(x = attn_weights_2_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input4_1_cast_fp16 = softmax(axis = var_18, x = input_5_cast_fp16)[name = tensor<string, []>("input4_1_cast_fp16")];
            tensor<bool, []> attn_output_2_transpose_x_0 = const()[name = tensor<string, []>("attn_output_2_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_2_transpose_y_0 = const()[name = tensor<string, []>("attn_output_2_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_2_cast_fp16 = transpose(perm = value_2_perm_0, x = var_127_cast_fp16)[name = tensor<string, []>("transpose_107")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_2_cast_fp16 = matmul(transpose_x = attn_output_2_transpose_x_0, transpose_y = attn_output_2_transpose_y_0, x = input4_1_cast_fp16, y = value_2_cast_fp16)[name = tensor<string, []>("attn_output_2_cast_fp16")];
            tensor<int32, [4]> var_137_perm_0 = const()[name = tensor<string, []>("op_137_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_139 = const()[name = tensor<string, []>("op_139"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_137_cast_fp16 = transpose(perm = var_137_perm_0, x = attn_output_2_cast_fp16)[name = tensor<string, []>("transpose_104")];
            tensor<fp16, [1, 256, 1024]> var_140_cast_fp16 = reshape(shape = var_139, x = var_137_cast_fp16)[name = tensor<string, []>("op_140_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533113536)))];
            tensor<fp16, [1024]> encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(535210752)))];
            tensor<fp16, [1, 256, 1024]> linear_3_cast_fp16 = linear(bias = encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_out_proj_weight_to_fp16, x = var_140_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input2_1_cast_fp16 = add(x = input0_7_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input2_1_cast_fp16")];
            tensor<int32, [1]> input0_5_axes_0 = const()[name = tensor<string, []>("input0_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_0_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(535212864)))];
            tensor<fp16, [1024]> encoder_layers_0_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(535214976)))];
            tensor<fp16, [1, 256, 1024]> input0_5_cast_fp16 = layer_norm(axes = input0_5_axes_0, beta = encoder_layers_0_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_0_final_layer_norm_weight_to_fp16, x = input2_1_cast_fp16)[name = tensor<string, []>("input0_5_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_0_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(535217088)))];
            tensor<fp16, [4096]> encoder_layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543605760)))];
            tensor<fp16, [1, 256, 4096]> linear_4_cast_fp16 = linear(bias = encoder_layers_0_fc1_bias_to_fp16, weight = encoder_layers_0_fc1_weight_to_fp16, x = input0_5_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_154_cast_fp16 = relu(x = linear_4_cast_fp16)[name = tensor<string, []>("op_154_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_0_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(543614016)))];
            tensor<fp16, [1024]> encoder_layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552002688)))];
            tensor<fp16, [1, 256, 1024]> linear_5_cast_fp16 = linear(bias = encoder_layers_0_fc2_bias_to_fp16, weight = encoder_layers_0_fc2_weight_to_fp16, x = var_154_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_160_cast_fp16 = add(x = input2_1_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("op_160_cast_fp16")];
            tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_1_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552004800)))];
            tensor<fp16, [1024]> encoder_layers_1_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552006912)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = encoder_layers_1_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_1_self_attn_layer_norm_weight_to_fp16, x = var_160_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552009024)))];
            tensor<fp16, [1024]> encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(554106240)))];
            tensor<fp16, [1, 256, 1024]> linear_6_cast_fp16 = linear(bias = encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_q_proj_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
            tensor<int32, [4]> var_179 = const()[name = tensor<string, []>("op_179"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_180_cast_fp16 = reshape(shape = var_179, x = linear_6_cast_fp16)[name = tensor<string, []>("op_180_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(554108352)))];
            tensor<fp16, [1024]> encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(556205568)))];
            tensor<fp16, [1, 256, 1024]> linear_7_cast_fp16 = linear(bias = encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_k_proj_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(556207680)))];
            tensor<fp16, [1024]> encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(558304896)))];
            tensor<fp16, [1, 256, 1024]> linear_8_cast_fp16 = linear(bias = encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_v_proj_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
            tensor<int32, [4]> var_188 = const()[name = tensor<string, []>("op_188"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_189_cast_fp16 = reshape(shape = var_188, x = linear_7_cast_fp16)[name = tensor<string, []>("op_189_cast_fp16")];
            tensor<int32, [4]> var_191 = const()[name = tensor<string, []>("op_191"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_192_cast_fp16 = reshape(shape = var_191, x = linear_8_cast_fp16)[name = tensor<string, []>("op_192_cast_fp16")];
            tensor<int32, [4]> value_4_perm_0 = const()[name = tensor<string, []>("value_4_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_195_transpose_x_0 = const()[name = tensor<string, []>("op_195_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_195_transpose_y_0 = const()[name = tensor<string, []>("op_195_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_38_perm_0 = const()[name = tensor<string, []>("transpose_38_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_39_perm_0 = const()[name = tensor<string, []>("transpose_39_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_39 = transpose(perm = transpose_39_perm_0, x = var_189_cast_fp16)[name = tensor<string, []>("transpose_101")];
            tensor<fp16, [1, 16, 256, 64]> transpose_38 = transpose(perm = transpose_38_perm_0, x = var_180_cast_fp16)[name = tensor<string, []>("transpose_102")];
            tensor<fp16, [1, 16, 256, 256]> var_195_cast_fp16 = matmul(transpose_x = var_195_transpose_x_0, transpose_y = var_195_transpose_y_0, x = transpose_38, y = transpose_39)[name = tensor<string, []>("op_195_cast_fp16")];
            tensor<fp16, []> var_196_to_fp16 = const()[name = tensor<string, []>("op_196_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_4_cast_fp16 = mul(x = var_195_cast_fp16, y = var_196_to_fp16)[name = tensor<string, []>("attn_weights_4_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_11_cast_fp16 = add(x = attn_weights_4_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_9_cast_fp16 = softmax(axis = var_18, x = input_11_cast_fp16)[name = tensor<string, []>("input0_9_cast_fp16")];
            tensor<bool, []> attn_output_4_transpose_x_0 = const()[name = tensor<string, []>("attn_output_4_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_4_transpose_y_0 = const()[name = tensor<string, []>("attn_output_4_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_4_cast_fp16 = transpose(perm = value_4_perm_0, x = var_192_cast_fp16)[name = tensor<string, []>("transpose_103")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_4_cast_fp16 = matmul(transpose_x = attn_output_4_transpose_x_0, transpose_y = attn_output_4_transpose_y_0, x = input0_9_cast_fp16, y = value_4_cast_fp16)[name = tensor<string, []>("attn_output_4_cast_fp16")];
            tensor<int32, [4]> var_202_perm_0 = const()[name = tensor<string, []>("op_202_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_204 = const()[name = tensor<string, []>("op_204"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_202_cast_fp16 = transpose(perm = var_202_perm_0, x = attn_output_4_cast_fp16)[name = tensor<string, []>("transpose_100")];
            tensor<fp16, [1, 256, 1024]> var_205_cast_fp16 = reshape(shape = var_204, x = var_202_cast_fp16)[name = tensor<string, []>("op_205_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(558307008)))];
            tensor<fp16, [1024]> encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560404224)))];
            tensor<fp16, [1, 256, 1024]> linear_9_cast_fp16 = linear(bias = encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_out_proj_weight_to_fp16, x = var_205_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_13_cast_fp16 = add(x = var_160_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<int32, [1]> input0_11_axes_0 = const()[name = tensor<string, []>("input0_11_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_1_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560406336)))];
            tensor<fp16, [1024]> encoder_layers_1_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560408448)))];
            tensor<fp16, [1, 256, 1024]> input0_11_cast_fp16 = layer_norm(axes = input0_11_axes_0, beta = encoder_layers_1_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_1_final_layer_norm_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("input0_11_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_1_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(560410560)))];
            tensor<fp16, [4096]> encoder_layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568799232)))];
            tensor<fp16, [1, 256, 4096]> linear_10_cast_fp16 = linear(bias = encoder_layers_1_fc1_bias_to_fp16, weight = encoder_layers_1_fc1_weight_to_fp16, x = input0_11_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_219_cast_fp16 = relu(x = linear_10_cast_fp16)[name = tensor<string, []>("op_219_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_1_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568807488)))];
            tensor<fp16, [1024]> encoder_layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577196160)))];
            tensor<fp16, [1, 256, 1024]> linear_11_cast_fp16 = linear(bias = encoder_layers_1_fc2_bias_to_fp16, weight = encoder_layers_1_fc2_weight_to_fp16, x = var_219_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_225_cast_fp16 = add(x = input_13_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("op_225_cast_fp16")];
            tensor<int32, [1]> hidden_states_11_axes_0 = const()[name = tensor<string, []>("hidden_states_11_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_2_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577198272)))];
            tensor<fp16, [1024]> encoder_layers_2_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577200384)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_11_cast_fp16 = layer_norm(axes = hidden_states_11_axes_0, beta = encoder_layers_2_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_2_self_attn_layer_norm_weight_to_fp16, x = var_225_cast_fp16)[name = tensor<string, []>("hidden_states_11_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577202496)))];
            tensor<fp16, [1024]> encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579299712)))];
            tensor<fp16, [1, 256, 1024]> linear_12_cast_fp16 = linear(bias = encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_q_proj_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<int32, [4]> var_244 = const()[name = tensor<string, []>("op_244"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_245_cast_fp16 = reshape(shape = var_244, x = linear_12_cast_fp16)[name = tensor<string, []>("op_245_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579301824)))];
            tensor<fp16, [1024]> encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(581399040)))];
            tensor<fp16, [1, 256, 1024]> linear_13_cast_fp16 = linear(bias = encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_k_proj_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(581401152)))];
            tensor<fp16, [1024]> encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(583498368)))];
            tensor<fp16, [1, 256, 1024]> linear_14_cast_fp16 = linear(bias = encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_v_proj_weight_to_fp16, x = hidden_states_11_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<int32, [4]> var_253 = const()[name = tensor<string, []>("op_253"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_254_cast_fp16 = reshape(shape = var_253, x = linear_13_cast_fp16)[name = tensor<string, []>("op_254_cast_fp16")];
            tensor<int32, [4]> var_256 = const()[name = tensor<string, []>("op_256"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_257_cast_fp16 = reshape(shape = var_256, x = linear_14_cast_fp16)[name = tensor<string, []>("op_257_cast_fp16")];
            tensor<int32, [4]> value_6_perm_0 = const()[name = tensor<string, []>("value_6_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_260_transpose_x_0 = const()[name = tensor<string, []>("op_260_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_260_transpose_y_0 = const()[name = tensor<string, []>("op_260_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_40_perm_0 = const()[name = tensor<string, []>("transpose_40_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_41_perm_0 = const()[name = tensor<string, []>("transpose_41_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_41 = transpose(perm = transpose_41_perm_0, x = var_254_cast_fp16)[name = tensor<string, []>("transpose_97")];
            tensor<fp16, [1, 16, 256, 64]> transpose_40 = transpose(perm = transpose_40_perm_0, x = var_245_cast_fp16)[name = tensor<string, []>("transpose_98")];
            tensor<fp16, [1, 16, 256, 256]> var_260_cast_fp16 = matmul(transpose_x = var_260_transpose_x_0, transpose_y = var_260_transpose_y_0, x = transpose_40, y = transpose_41)[name = tensor<string, []>("op_260_cast_fp16")];
            tensor<fp16, []> var_261_to_fp16 = const()[name = tensor<string, []>("op_261_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_6_cast_fp16 = mul(x = var_260_cast_fp16, y = var_261_to_fp16)[name = tensor<string, []>("attn_weights_6_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_17_cast_fp16 = add(x = attn_weights_6_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_15_cast_fp16 = softmax(axis = var_18, x = input_17_cast_fp16)[name = tensor<string, []>("input0_15_cast_fp16")];
            tensor<bool, []> attn_output_6_transpose_x_0 = const()[name = tensor<string, []>("attn_output_6_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_6_transpose_y_0 = const()[name = tensor<string, []>("attn_output_6_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_6_cast_fp16 = transpose(perm = value_6_perm_0, x = var_257_cast_fp16)[name = tensor<string, []>("transpose_99")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_6_cast_fp16 = matmul(transpose_x = attn_output_6_transpose_x_0, transpose_y = attn_output_6_transpose_y_0, x = input0_15_cast_fp16, y = value_6_cast_fp16)[name = tensor<string, []>("attn_output_6_cast_fp16")];
            tensor<int32, [4]> var_267_perm_0 = const()[name = tensor<string, []>("op_267_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_269 = const()[name = tensor<string, []>("op_269"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_267_cast_fp16 = transpose(perm = var_267_perm_0, x = attn_output_6_cast_fp16)[name = tensor<string, []>("transpose_96")];
            tensor<fp16, [1, 256, 1024]> var_270_cast_fp16 = reshape(shape = var_269, x = var_267_cast_fp16)[name = tensor<string, []>("op_270_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(583500480)))];
            tensor<fp16, [1024]> encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585597696)))];
            tensor<fp16, [1, 256, 1024]> linear_15_cast_fp16 = linear(bias = encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_out_proj_weight_to_fp16, x = var_270_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_19_cast_fp16 = add(x = var_225_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
            tensor<int32, [1]> input0_17_axes_0 = const()[name = tensor<string, []>("input0_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_2_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585599808)))];
            tensor<fp16, [1024]> encoder_layers_2_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585601920)))];
            tensor<fp16, [1, 256, 1024]> input0_17_cast_fp16 = layer_norm(axes = input0_17_axes_0, beta = encoder_layers_2_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_2_final_layer_norm_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("input0_17_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_2_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585604032)))];
            tensor<fp16, [4096]> encoder_layers_2_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(593992704)))];
            tensor<fp16, [1, 256, 4096]> linear_16_cast_fp16 = linear(bias = encoder_layers_2_fc1_bias_to_fp16, weight = encoder_layers_2_fc1_weight_to_fp16, x = input0_17_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_284_cast_fp16 = relu(x = linear_16_cast_fp16)[name = tensor<string, []>("op_284_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_2_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(594000960)))];
            tensor<fp16, [1024]> encoder_layers_2_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602389632)))];
            tensor<fp16, [1, 256, 1024]> linear_17_cast_fp16 = linear(bias = encoder_layers_2_fc2_bias_to_fp16, weight = encoder_layers_2_fc2_weight_to_fp16, x = var_284_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_290_cast_fp16 = add(x = input_19_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("op_290_cast_fp16")];
            tensor<int32, [1]> hidden_states_15_axes_0 = const()[name = tensor<string, []>("hidden_states_15_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_3_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602391744)))];
            tensor<fp16, [1024]> encoder_layers_3_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602393856)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_15_cast_fp16 = layer_norm(axes = hidden_states_15_axes_0, beta = encoder_layers_3_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_3_self_attn_layer_norm_weight_to_fp16, x = var_290_cast_fp16)[name = tensor<string, []>("hidden_states_15_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602395968)))];
            tensor<fp16, [1024]> encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604493184)))];
            tensor<fp16, [1, 256, 1024]> linear_18_cast_fp16 = linear(bias = encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_q_proj_weight_to_fp16, x = hidden_states_15_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<int32, [4]> var_309 = const()[name = tensor<string, []>("op_309"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_310_cast_fp16 = reshape(shape = var_309, x = linear_18_cast_fp16)[name = tensor<string, []>("op_310_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604495296)))];
            tensor<fp16, [1024]> encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606592512)))];
            tensor<fp16, [1, 256, 1024]> linear_19_cast_fp16 = linear(bias = encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_k_proj_weight_to_fp16, x = hidden_states_15_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(606594624)))];
            tensor<fp16, [1024]> encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608691840)))];
            tensor<fp16, [1, 256, 1024]> linear_20_cast_fp16 = linear(bias = encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_v_proj_weight_to_fp16, x = hidden_states_15_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<int32, [4]> var_318 = const()[name = tensor<string, []>("op_318"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_319_cast_fp16 = reshape(shape = var_318, x = linear_19_cast_fp16)[name = tensor<string, []>("op_319_cast_fp16")];
            tensor<int32, [4]> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_322_cast_fp16 = reshape(shape = var_321, x = linear_20_cast_fp16)[name = tensor<string, []>("op_322_cast_fp16")];
            tensor<int32, [4]> value_8_perm_0 = const()[name = tensor<string, []>("value_8_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_325_transpose_x_0 = const()[name = tensor<string, []>("op_325_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_325_transpose_y_0 = const()[name = tensor<string, []>("op_325_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_42_perm_0 = const()[name = tensor<string, []>("transpose_42_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_43_perm_0 = const()[name = tensor<string, []>("transpose_43_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_43 = transpose(perm = transpose_43_perm_0, x = var_319_cast_fp16)[name = tensor<string, []>("transpose_93")];
            tensor<fp16, [1, 16, 256, 64]> transpose_42 = transpose(perm = transpose_42_perm_0, x = var_310_cast_fp16)[name = tensor<string, []>("transpose_94")];
            tensor<fp16, [1, 16, 256, 256]> var_325_cast_fp16 = matmul(transpose_x = var_325_transpose_x_0, transpose_y = var_325_transpose_y_0, x = transpose_42, y = transpose_43)[name = tensor<string, []>("op_325_cast_fp16")];
            tensor<fp16, []> var_326_to_fp16 = const()[name = tensor<string, []>("op_326_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_8_cast_fp16 = mul(x = var_325_cast_fp16, y = var_326_to_fp16)[name = tensor<string, []>("attn_weights_8_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_23_cast_fp16 = add(x = attn_weights_8_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_21_cast_fp16 = softmax(axis = var_18, x = input_23_cast_fp16)[name = tensor<string, []>("input0_21_cast_fp16")];
            tensor<bool, []> attn_output_8_transpose_x_0 = const()[name = tensor<string, []>("attn_output_8_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_8_transpose_y_0 = const()[name = tensor<string, []>("attn_output_8_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_8_cast_fp16 = transpose(perm = value_8_perm_0, x = var_322_cast_fp16)[name = tensor<string, []>("transpose_95")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_8_cast_fp16 = matmul(transpose_x = attn_output_8_transpose_x_0, transpose_y = attn_output_8_transpose_y_0, x = input0_21_cast_fp16, y = value_8_cast_fp16)[name = tensor<string, []>("attn_output_8_cast_fp16")];
            tensor<int32, [4]> var_332_perm_0 = const()[name = tensor<string, []>("op_332_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_334 = const()[name = tensor<string, []>("op_334"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_332_cast_fp16 = transpose(perm = var_332_perm_0, x = attn_output_8_cast_fp16)[name = tensor<string, []>("transpose_92")];
            tensor<fp16, [1, 256, 1024]> var_335_cast_fp16 = reshape(shape = var_334, x = var_332_cast_fp16)[name = tensor<string, []>("op_335_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608693952)))];
            tensor<fp16, [1024]> encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(610791168)))];
            tensor<fp16, [1, 256, 1024]> linear_21_cast_fp16 = linear(bias = encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_out_proj_weight_to_fp16, x = var_335_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_25_cast_fp16 = add(x = var_290_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<int32, [1]> input0_23_axes_0 = const()[name = tensor<string, []>("input0_23_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_3_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(610793280)))];
            tensor<fp16, [1024]> encoder_layers_3_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(610795392)))];
            tensor<fp16, [1, 256, 1024]> input0_23_cast_fp16 = layer_norm(axes = input0_23_axes_0, beta = encoder_layers_3_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_3_final_layer_norm_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("input0_23_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_3_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(610797504)))];
            tensor<fp16, [4096]> encoder_layers_3_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(619186176)))];
            tensor<fp16, [1, 256, 4096]> linear_22_cast_fp16 = linear(bias = encoder_layers_3_fc1_bias_to_fp16, weight = encoder_layers_3_fc1_weight_to_fp16, x = input0_23_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_349_cast_fp16 = relu(x = linear_22_cast_fp16)[name = tensor<string, []>("op_349_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_3_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(619194432)))];
            tensor<fp16, [1024]> encoder_layers_3_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(627583104)))];
            tensor<fp16, [1, 256, 1024]> linear_23_cast_fp16 = linear(bias = encoder_layers_3_fc2_bias_to_fp16, weight = encoder_layers_3_fc2_weight_to_fp16, x = var_349_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_355_cast_fp16 = add(x = input_25_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("op_355_cast_fp16")];
            tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = tensor<string, []>("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_4_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(627585216)))];
            tensor<fp16, [1024]> encoder_layers_4_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(627587328)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = encoder_layers_4_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_4_self_attn_layer_norm_weight_to_fp16, x = var_355_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(627589440)))];
            tensor<fp16, [1024]> encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(629686656)))];
            tensor<fp16, [1, 256, 1024]> linear_24_cast_fp16 = linear(bias = encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_q_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")];
            tensor<int32, [4]> var_374 = const()[name = tensor<string, []>("op_374"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_375_cast_fp16 = reshape(shape = var_374, x = linear_24_cast_fp16)[name = tensor<string, []>("op_375_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(629688768)))];
            tensor<fp16, [1024]> encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631785984)))];
            tensor<fp16, [1, 256, 1024]> linear_25_cast_fp16 = linear(bias = encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_k_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(631788096)))];
            tensor<fp16, [1024]> encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633885312)))];
            tensor<fp16, [1, 256, 1024]> linear_26_cast_fp16 = linear(bias = encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_v_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<int32, [4]> var_383 = const()[name = tensor<string, []>("op_383"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_384_cast_fp16 = reshape(shape = var_383, x = linear_25_cast_fp16)[name = tensor<string, []>("op_384_cast_fp16")];
            tensor<int32, [4]> var_386 = const()[name = tensor<string, []>("op_386"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_387_cast_fp16 = reshape(shape = var_386, x = linear_26_cast_fp16)[name = tensor<string, []>("op_387_cast_fp16")];
            tensor<int32, [4]> value_10_perm_0 = const()[name = tensor<string, []>("value_10_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_390_transpose_x_0 = const()[name = tensor<string, []>("op_390_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_390_transpose_y_0 = const()[name = tensor<string, []>("op_390_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_44_perm_0 = const()[name = tensor<string, []>("transpose_44_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_45_perm_0 = const()[name = tensor<string, []>("transpose_45_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_45 = transpose(perm = transpose_45_perm_0, x = var_384_cast_fp16)[name = tensor<string, []>("transpose_89")];
            tensor<fp16, [1, 16, 256, 64]> transpose_44 = transpose(perm = transpose_44_perm_0, x = var_375_cast_fp16)[name = tensor<string, []>("transpose_90")];
            tensor<fp16, [1, 16, 256, 256]> var_390_cast_fp16 = matmul(transpose_x = var_390_transpose_x_0, transpose_y = var_390_transpose_y_0, x = transpose_44, y = transpose_45)[name = tensor<string, []>("op_390_cast_fp16")];
            tensor<fp16, []> var_391_to_fp16 = const()[name = tensor<string, []>("op_391_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_10_cast_fp16 = mul(x = var_390_cast_fp16, y = var_391_to_fp16)[name = tensor<string, []>("attn_weights_10_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_29_cast_fp16 = add(x = attn_weights_10_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_27_cast_fp16 = softmax(axis = var_18, x = input_29_cast_fp16)[name = tensor<string, []>("input0_27_cast_fp16")];
            tensor<bool, []> attn_output_10_transpose_x_0 = const()[name = tensor<string, []>("attn_output_10_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_10_transpose_y_0 = const()[name = tensor<string, []>("attn_output_10_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_10_cast_fp16 = transpose(perm = value_10_perm_0, x = var_387_cast_fp16)[name = tensor<string, []>("transpose_91")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_10_cast_fp16 = matmul(transpose_x = attn_output_10_transpose_x_0, transpose_y = attn_output_10_transpose_y_0, x = input0_27_cast_fp16, y = value_10_cast_fp16)[name = tensor<string, []>("attn_output_10_cast_fp16")];
            tensor<int32, [4]> var_397_perm_0 = const()[name = tensor<string, []>("op_397_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_399 = const()[name = tensor<string, []>("op_399"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_397_cast_fp16 = transpose(perm = var_397_perm_0, x = attn_output_10_cast_fp16)[name = tensor<string, []>("transpose_88")];
            tensor<fp16, [1, 256, 1024]> var_400_cast_fp16 = reshape(shape = var_399, x = var_397_cast_fp16)[name = tensor<string, []>("op_400_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633887424)))];
            tensor<fp16, [1024]> encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635984640)))];
            tensor<fp16, [1, 256, 1024]> linear_27_cast_fp16 = linear(bias = encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_out_proj_weight_to_fp16, x = var_400_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_31_cast_fp16 = add(x = var_355_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
            tensor<int32, [1]> input0_29_axes_0 = const()[name = tensor<string, []>("input0_29_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_4_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635986752)))];
            tensor<fp16, [1024]> encoder_layers_4_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635988864)))];
            tensor<fp16, [1, 256, 1024]> input0_29_cast_fp16 = layer_norm(axes = input0_29_axes_0, beta = encoder_layers_4_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_4_final_layer_norm_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("input0_29_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_4_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(635990976)))];
            tensor<fp16, [4096]> encoder_layers_4_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644379648)))];
            tensor<fp16, [1, 256, 4096]> linear_28_cast_fp16 = linear(bias = encoder_layers_4_fc1_bias_to_fp16, weight = encoder_layers_4_fc1_weight_to_fp16, x = input0_29_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_414_cast_fp16 = relu(x = linear_28_cast_fp16)[name = tensor<string, []>("op_414_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_4_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644387904)))];
            tensor<fp16, [1024]> encoder_layers_4_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652776576)))];
            tensor<fp16, [1, 256, 1024]> linear_29_cast_fp16 = linear(bias = encoder_layers_4_fc2_bias_to_fp16, weight = encoder_layers_4_fc2_weight_to_fp16, x = var_414_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_420_cast_fp16 = add(x = input_31_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("op_420_cast_fp16")];
            tensor<int32, [1]> hidden_states_23_axes_0 = const()[name = tensor<string, []>("hidden_states_23_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_5_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652778688)))];
            tensor<fp16, [1024]> encoder_layers_5_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652780800)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_23_cast_fp16 = layer_norm(axes = hidden_states_23_axes_0, beta = encoder_layers_5_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_5_self_attn_layer_norm_weight_to_fp16, x = var_420_cast_fp16)[name = tensor<string, []>("hidden_states_23_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652782912)))];
            tensor<fp16, [1024]> encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(654880128)))];
            tensor<fp16, [1, 256, 1024]> linear_30_cast_fp16 = linear(bias = encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_q_proj_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
            tensor<int32, [4]> var_439 = const()[name = tensor<string, []>("op_439"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_440_cast_fp16 = reshape(shape = var_439, x = linear_30_cast_fp16)[name = tensor<string, []>("op_440_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(654882240)))];
            tensor<fp16, [1024]> encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(656979456)))];
            tensor<fp16, [1, 256, 1024]> linear_31_cast_fp16 = linear(bias = encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_k_proj_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(656981568)))];
            tensor<fp16, [1024]> encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(659078784)))];
            tensor<fp16, [1, 256, 1024]> linear_32_cast_fp16 = linear(bias = encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_v_proj_weight_to_fp16, x = hidden_states_23_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<int32, [4]> var_448 = const()[name = tensor<string, []>("op_448"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_449_cast_fp16 = reshape(shape = var_448, x = linear_31_cast_fp16)[name = tensor<string, []>("op_449_cast_fp16")];
            tensor<int32, [4]> var_451 = const()[name = tensor<string, []>("op_451"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_452_cast_fp16 = reshape(shape = var_451, x = linear_32_cast_fp16)[name = tensor<string, []>("op_452_cast_fp16")];
            tensor<int32, [4]> value_12_perm_0 = const()[name = tensor<string, []>("value_12_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_455_transpose_x_0 = const()[name = tensor<string, []>("op_455_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_455_transpose_y_0 = const()[name = tensor<string, []>("op_455_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_46_perm_0 = const()[name = tensor<string, []>("transpose_46_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_47_perm_0 = const()[name = tensor<string, []>("transpose_47_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_47 = transpose(perm = transpose_47_perm_0, x = var_449_cast_fp16)[name = tensor<string, []>("transpose_85")];
            tensor<fp16, [1, 16, 256, 64]> transpose_46 = transpose(perm = transpose_46_perm_0, x = var_440_cast_fp16)[name = tensor<string, []>("transpose_86")];
            tensor<fp16, [1, 16, 256, 256]> var_455_cast_fp16 = matmul(transpose_x = var_455_transpose_x_0, transpose_y = var_455_transpose_y_0, x = transpose_46, y = transpose_47)[name = tensor<string, []>("op_455_cast_fp16")];
            tensor<fp16, []> var_456_to_fp16 = const()[name = tensor<string, []>("op_456_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_12_cast_fp16 = mul(x = var_455_cast_fp16, y = var_456_to_fp16)[name = tensor<string, []>("attn_weights_12_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_35_cast_fp16 = add(x = attn_weights_12_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_33_cast_fp16 = softmax(axis = var_18, x = input_35_cast_fp16)[name = tensor<string, []>("input0_33_cast_fp16")];
            tensor<bool, []> attn_output_12_transpose_x_0 = const()[name = tensor<string, []>("attn_output_12_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_12_transpose_y_0 = const()[name = tensor<string, []>("attn_output_12_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_12_cast_fp16 = transpose(perm = value_12_perm_0, x = var_452_cast_fp16)[name = tensor<string, []>("transpose_87")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_12_cast_fp16 = matmul(transpose_x = attn_output_12_transpose_x_0, transpose_y = attn_output_12_transpose_y_0, x = input0_33_cast_fp16, y = value_12_cast_fp16)[name = tensor<string, []>("attn_output_12_cast_fp16")];
            tensor<int32, [4]> var_462_perm_0 = const()[name = tensor<string, []>("op_462_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_464 = const()[name = tensor<string, []>("op_464"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_462_cast_fp16 = transpose(perm = var_462_perm_0, x = attn_output_12_cast_fp16)[name = tensor<string, []>("transpose_84")];
            tensor<fp16, [1, 256, 1024]> var_465_cast_fp16 = reshape(shape = var_464, x = var_462_cast_fp16)[name = tensor<string, []>("op_465_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(659080896)))];
            tensor<fp16, [1024]> encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661178112)))];
            tensor<fp16, [1, 256, 1024]> linear_33_cast_fp16 = linear(bias = encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_out_proj_weight_to_fp16, x = var_465_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_37_cast_fp16 = add(x = var_420_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")];
            tensor<int32, [1]> input0_35_axes_0 = const()[name = tensor<string, []>("input0_35_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_5_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661180224)))];
            tensor<fp16, [1024]> encoder_layers_5_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661182336)))];
            tensor<fp16, [1, 256, 1024]> input0_35_cast_fp16 = layer_norm(axes = input0_35_axes_0, beta = encoder_layers_5_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_5_final_layer_norm_weight_to_fp16, x = input_37_cast_fp16)[name = tensor<string, []>("input0_35_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_5_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661184448)))];
            tensor<fp16, [4096]> encoder_layers_5_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(669573120)))];
            tensor<fp16, [1, 256, 4096]> linear_34_cast_fp16 = linear(bias = encoder_layers_5_fc1_bias_to_fp16, weight = encoder_layers_5_fc1_weight_to_fp16, x = input0_35_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_479_cast_fp16 = relu(x = linear_34_cast_fp16)[name = tensor<string, []>("op_479_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_5_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(669581376)))];
            tensor<fp16, [1024]> encoder_layers_5_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677970048)))];
            tensor<fp16, [1, 256, 1024]> linear_35_cast_fp16 = linear(bias = encoder_layers_5_fc2_bias_to_fp16, weight = encoder_layers_5_fc2_weight_to_fp16, x = var_479_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_485_cast_fp16 = add(x = input_37_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("op_485_cast_fp16")];
            tensor<int32, [1]> hidden_states_27_axes_0 = const()[name = tensor<string, []>("hidden_states_27_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_6_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677972160)))];
            tensor<fp16, [1024]> encoder_layers_6_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677974272)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_27_cast_fp16 = layer_norm(axes = hidden_states_27_axes_0, beta = encoder_layers_6_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_6_self_attn_layer_norm_weight_to_fp16, x = var_485_cast_fp16)[name = tensor<string, []>("hidden_states_27_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677976384)))];
            tensor<fp16, [1024]> encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(680073600)))];
            tensor<fp16, [1, 256, 1024]> linear_36_cast_fp16 = linear(bias = encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_q_proj_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<int32, [4]> var_504 = const()[name = tensor<string, []>("op_504"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_505_cast_fp16 = reshape(shape = var_504, x = linear_36_cast_fp16)[name = tensor<string, []>("op_505_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(680075712)))];
            tensor<fp16, [1024]> encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(682172928)))];
            tensor<fp16, [1, 256, 1024]> linear_37_cast_fp16 = linear(bias = encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_k_proj_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(682175040)))];
            tensor<fp16, [1024]> encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684272256)))];
            tensor<fp16, [1, 256, 1024]> linear_38_cast_fp16 = linear(bias = encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_v_proj_weight_to_fp16, x = hidden_states_27_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<int32, [4]> var_513 = const()[name = tensor<string, []>("op_513"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_514_cast_fp16 = reshape(shape = var_513, x = linear_37_cast_fp16)[name = tensor<string, []>("op_514_cast_fp16")];
            tensor<int32, [4]> var_516 = const()[name = tensor<string, []>("op_516"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_517_cast_fp16 = reshape(shape = var_516, x = linear_38_cast_fp16)[name = tensor<string, []>("op_517_cast_fp16")];
            tensor<int32, [4]> value_14_perm_0 = const()[name = tensor<string, []>("value_14_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_520_transpose_x_0 = const()[name = tensor<string, []>("op_520_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_520_transpose_y_0 = const()[name = tensor<string, []>("op_520_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_48_perm_0 = const()[name = tensor<string, []>("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_49_perm_0 = const()[name = tensor<string, []>("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_49 = transpose(perm = transpose_49_perm_0, x = var_514_cast_fp16)[name = tensor<string, []>("transpose_81")];
            tensor<fp16, [1, 16, 256, 64]> transpose_48 = transpose(perm = transpose_48_perm_0, x = var_505_cast_fp16)[name = tensor<string, []>("transpose_82")];
            tensor<fp16, [1, 16, 256, 256]> var_520_cast_fp16 = matmul(transpose_x = var_520_transpose_x_0, transpose_y = var_520_transpose_y_0, x = transpose_48, y = transpose_49)[name = tensor<string, []>("op_520_cast_fp16")];
            tensor<fp16, []> var_521_to_fp16 = const()[name = tensor<string, []>("op_521_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_14_cast_fp16 = mul(x = var_520_cast_fp16, y = var_521_to_fp16)[name = tensor<string, []>("attn_weights_14_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_41_cast_fp16 = add(x = attn_weights_14_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_39_cast_fp16 = softmax(axis = var_18, x = input_41_cast_fp16)[name = tensor<string, []>("input0_39_cast_fp16")];
            tensor<bool, []> attn_output_14_transpose_x_0 = const()[name = tensor<string, []>("attn_output_14_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_14_transpose_y_0 = const()[name = tensor<string, []>("attn_output_14_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_14_cast_fp16 = transpose(perm = value_14_perm_0, x = var_517_cast_fp16)[name = tensor<string, []>("transpose_83")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_14_cast_fp16 = matmul(transpose_x = attn_output_14_transpose_x_0, transpose_y = attn_output_14_transpose_y_0, x = input0_39_cast_fp16, y = value_14_cast_fp16)[name = tensor<string, []>("attn_output_14_cast_fp16")];
            tensor<int32, [4]> var_527_perm_0 = const()[name = tensor<string, []>("op_527_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_529 = const()[name = tensor<string, []>("op_529"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_527_cast_fp16 = transpose(perm = var_527_perm_0, x = attn_output_14_cast_fp16)[name = tensor<string, []>("transpose_80")];
            tensor<fp16, [1, 256, 1024]> var_530_cast_fp16 = reshape(shape = var_529, x = var_527_cast_fp16)[name = tensor<string, []>("op_530_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684274368)))];
            tensor<fp16, [1024]> encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686371584)))];
            tensor<fp16, [1, 256, 1024]> linear_39_cast_fp16 = linear(bias = encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_out_proj_weight_to_fp16, x = var_530_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_43_cast_fp16 = add(x = var_485_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")];
            tensor<int32, [1]> input0_41_axes_0 = const()[name = tensor<string, []>("input0_41_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_6_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686373696)))];
            tensor<fp16, [1024]> encoder_layers_6_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686375808)))];
            tensor<fp16, [1, 256, 1024]> input0_41_cast_fp16 = layer_norm(axes = input0_41_axes_0, beta = encoder_layers_6_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_6_final_layer_norm_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("input0_41_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_6_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686377920)))];
            tensor<fp16, [4096]> encoder_layers_6_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(694766592)))];
            tensor<fp16, [1, 256, 4096]> linear_40_cast_fp16 = linear(bias = encoder_layers_6_fc1_bias_to_fp16, weight = encoder_layers_6_fc1_weight_to_fp16, x = input0_41_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_544_cast_fp16 = relu(x = linear_40_cast_fp16)[name = tensor<string, []>("op_544_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_6_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(694774848)))];
            tensor<fp16, [1024]> encoder_layers_6_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703163520)))];
            tensor<fp16, [1, 256, 1024]> linear_41_cast_fp16 = linear(bias = encoder_layers_6_fc2_bias_to_fp16, weight = encoder_layers_6_fc2_weight_to_fp16, x = var_544_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_550_cast_fp16 = add(x = input_43_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("op_550_cast_fp16")];
            tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_7_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703165632)))];
            tensor<fp16, [1024]> encoder_layers_7_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703167744)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = encoder_layers_7_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_7_self_attn_layer_norm_weight_to_fp16, x = var_550_cast_fp16)[name = tensor<string, []>("hidden_states_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703169856)))];
            tensor<fp16, [1024]> encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(705267072)))];
            tensor<fp16, [1, 256, 1024]> linear_42_cast_fp16 = linear(bias = encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_q_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")];
            tensor<int32, [4]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_570_cast_fp16 = reshape(shape = var_569, x = linear_42_cast_fp16)[name = tensor<string, []>("op_570_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(705269184)))];
            tensor<fp16, [1024]> encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(707366400)))];
            tensor<fp16, [1, 256, 1024]> linear_43_cast_fp16 = linear(bias = encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_k_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(707368512)))];
            tensor<fp16, [1024]> encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709465728)))];
            tensor<fp16, [1, 256, 1024]> linear_44_cast_fp16 = linear(bias = encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_v_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<int32, [4]> var_578 = const()[name = tensor<string, []>("op_578"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_579_cast_fp16 = reshape(shape = var_578, x = linear_43_cast_fp16)[name = tensor<string, []>("op_579_cast_fp16")];
            tensor<int32, [4]> var_581 = const()[name = tensor<string, []>("op_581"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_582_cast_fp16 = reshape(shape = var_581, x = linear_44_cast_fp16)[name = tensor<string, []>("op_582_cast_fp16")];
            tensor<int32, [4]> value_16_perm_0 = const()[name = tensor<string, []>("value_16_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_585_transpose_x_0 = const()[name = tensor<string, []>("op_585_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_585_transpose_y_0 = const()[name = tensor<string, []>("op_585_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_50_perm_0 = const()[name = tensor<string, []>("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_51_perm_0 = const()[name = tensor<string, []>("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_51 = transpose(perm = transpose_51_perm_0, x = var_579_cast_fp16)[name = tensor<string, []>("transpose_77")];
            tensor<fp16, [1, 16, 256, 64]> transpose_50 = transpose(perm = transpose_50_perm_0, x = var_570_cast_fp16)[name = tensor<string, []>("transpose_78")];
            tensor<fp16, [1, 16, 256, 256]> var_585_cast_fp16 = matmul(transpose_x = var_585_transpose_x_0, transpose_y = var_585_transpose_y_0, x = transpose_50, y = transpose_51)[name = tensor<string, []>("op_585_cast_fp16")];
            tensor<fp16, []> var_586_to_fp16 = const()[name = tensor<string, []>("op_586_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_16_cast_fp16 = mul(x = var_585_cast_fp16, y = var_586_to_fp16)[name = tensor<string, []>("attn_weights_16_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_47_cast_fp16 = add(x = attn_weights_16_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_45_cast_fp16 = softmax(axis = var_18, x = input_47_cast_fp16)[name = tensor<string, []>("input0_45_cast_fp16")];
            tensor<bool, []> attn_output_16_transpose_x_0 = const()[name = tensor<string, []>("attn_output_16_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_16_transpose_y_0 = const()[name = tensor<string, []>("attn_output_16_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_16_cast_fp16 = transpose(perm = value_16_perm_0, x = var_582_cast_fp16)[name = tensor<string, []>("transpose_79")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_16_cast_fp16 = matmul(transpose_x = attn_output_16_transpose_x_0, transpose_y = attn_output_16_transpose_y_0, x = input0_45_cast_fp16, y = value_16_cast_fp16)[name = tensor<string, []>("attn_output_16_cast_fp16")];
            tensor<int32, [4]> var_592_perm_0 = const()[name = tensor<string, []>("op_592_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_594 = const()[name = tensor<string, []>("op_594"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_592_cast_fp16 = transpose(perm = var_592_perm_0, x = attn_output_16_cast_fp16)[name = tensor<string, []>("transpose_76")];
            tensor<fp16, [1, 256, 1024]> var_595_cast_fp16 = reshape(shape = var_594, x = var_592_cast_fp16)[name = tensor<string, []>("op_595_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709467840)))];
            tensor<fp16, [1024]> encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711565056)))];
            tensor<fp16, [1, 256, 1024]> linear_45_cast_fp16 = linear(bias = encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_out_proj_weight_to_fp16, x = var_595_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_49_cast_fp16 = add(x = var_550_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")];
            tensor<int32, [1]> input0_47_axes_0 = const()[name = tensor<string, []>("input0_47_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_7_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711567168)))];
            tensor<fp16, [1024]> encoder_layers_7_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711569280)))];
            tensor<fp16, [1, 256, 1024]> input0_47_cast_fp16 = layer_norm(axes = input0_47_axes_0, beta = encoder_layers_7_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_7_final_layer_norm_weight_to_fp16, x = input_49_cast_fp16)[name = tensor<string, []>("input0_47_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_7_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711571392)))];
            tensor<fp16, [4096]> encoder_layers_7_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(719960064)))];
            tensor<fp16, [1, 256, 4096]> linear_46_cast_fp16 = linear(bias = encoder_layers_7_fc1_bias_to_fp16, weight = encoder_layers_7_fc1_weight_to_fp16, x = input0_47_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_609_cast_fp16 = relu(x = linear_46_cast_fp16)[name = tensor<string, []>("op_609_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_7_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(719968320)))];
            tensor<fp16, [1024]> encoder_layers_7_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728356992)))];
            tensor<fp16, [1, 256, 1024]> linear_47_cast_fp16 = linear(bias = encoder_layers_7_fc2_bias_to_fp16, weight = encoder_layers_7_fc2_weight_to_fp16, x = var_609_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_615_cast_fp16 = add(x = input_49_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("op_615_cast_fp16")];
            tensor<int32, [1]> hidden_states_35_axes_0 = const()[name = tensor<string, []>("hidden_states_35_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_8_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728359104)))];
            tensor<fp16, [1024]> encoder_layers_8_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728361216)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_35_cast_fp16 = layer_norm(axes = hidden_states_35_axes_0, beta = encoder_layers_8_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_8_self_attn_layer_norm_weight_to_fp16, x = var_615_cast_fp16)[name = tensor<string, []>("hidden_states_35_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(728363328)))];
            tensor<fp16, [1024]> encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(730460544)))];
            tensor<fp16, [1, 256, 1024]> linear_48_cast_fp16 = linear(bias = encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_q_proj_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<int32, [4]> var_634 = const()[name = tensor<string, []>("op_634"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_635_cast_fp16 = reshape(shape = var_634, x = linear_48_cast_fp16)[name = tensor<string, []>("op_635_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(730462656)))];
            tensor<fp16, [1024]> encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732559872)))];
            tensor<fp16, [1, 256, 1024]> linear_49_cast_fp16 = linear(bias = encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_k_proj_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732561984)))];
            tensor<fp16, [1024]> encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734659200)))];
            tensor<fp16, [1, 256, 1024]> linear_50_cast_fp16 = linear(bias = encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_v_proj_weight_to_fp16, x = hidden_states_35_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> var_643 = const()[name = tensor<string, []>("op_643"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_644_cast_fp16 = reshape(shape = var_643, x = linear_49_cast_fp16)[name = tensor<string, []>("op_644_cast_fp16")];
            tensor<int32, [4]> var_646 = const()[name = tensor<string, []>("op_646"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_647_cast_fp16 = reshape(shape = var_646, x = linear_50_cast_fp16)[name = tensor<string, []>("op_647_cast_fp16")];
            tensor<int32, [4]> value_18_perm_0 = const()[name = tensor<string, []>("value_18_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_650_transpose_x_0 = const()[name = tensor<string, []>("op_650_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_650_transpose_y_0 = const()[name = tensor<string, []>("op_650_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_52_perm_0 = const()[name = tensor<string, []>("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_53_perm_0 = const()[name = tensor<string, []>("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_53 = transpose(perm = transpose_53_perm_0, x = var_644_cast_fp16)[name = tensor<string, []>("transpose_73")];
            tensor<fp16, [1, 16, 256, 64]> transpose_52 = transpose(perm = transpose_52_perm_0, x = var_635_cast_fp16)[name = tensor<string, []>("transpose_74")];
            tensor<fp16, [1, 16, 256, 256]> var_650_cast_fp16 = matmul(transpose_x = var_650_transpose_x_0, transpose_y = var_650_transpose_y_0, x = transpose_52, y = transpose_53)[name = tensor<string, []>("op_650_cast_fp16")];
            tensor<fp16, []> var_651_to_fp16 = const()[name = tensor<string, []>("op_651_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_18_cast_fp16 = mul(x = var_650_cast_fp16, y = var_651_to_fp16)[name = tensor<string, []>("attn_weights_18_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_53_cast_fp16 = add(x = attn_weights_18_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_51_cast_fp16 = softmax(axis = var_18, x = input_53_cast_fp16)[name = tensor<string, []>("input0_51_cast_fp16")];
            tensor<bool, []> attn_output_18_transpose_x_0 = const()[name = tensor<string, []>("attn_output_18_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_18_transpose_y_0 = const()[name = tensor<string, []>("attn_output_18_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_18_cast_fp16 = transpose(perm = value_18_perm_0, x = var_647_cast_fp16)[name = tensor<string, []>("transpose_75")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_18_cast_fp16 = matmul(transpose_x = attn_output_18_transpose_x_0, transpose_y = attn_output_18_transpose_y_0, x = input0_51_cast_fp16, y = value_18_cast_fp16)[name = tensor<string, []>("attn_output_18_cast_fp16")];
            tensor<int32, [4]> var_657_perm_0 = const()[name = tensor<string, []>("op_657_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_659 = const()[name = tensor<string, []>("op_659"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_657_cast_fp16 = transpose(perm = var_657_perm_0, x = attn_output_18_cast_fp16)[name = tensor<string, []>("transpose_72")];
            tensor<fp16, [1, 256, 1024]> var_660_cast_fp16 = reshape(shape = var_659, x = var_657_cast_fp16)[name = tensor<string, []>("op_660_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734661312)))];
            tensor<fp16, [1024]> encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736758528)))];
            tensor<fp16, [1, 256, 1024]> linear_51_cast_fp16 = linear(bias = encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_out_proj_weight_to_fp16, x = var_660_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_55_cast_fp16 = add(x = var_615_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
            tensor<int32, [1]> input0_53_axes_0 = const()[name = tensor<string, []>("input0_53_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_8_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736760640)))];
            tensor<fp16, [1024]> encoder_layers_8_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736762752)))];
            tensor<fp16, [1, 256, 1024]> input0_53_cast_fp16 = layer_norm(axes = input0_53_axes_0, beta = encoder_layers_8_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_8_final_layer_norm_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("input0_53_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_8_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736764864)))];
            tensor<fp16, [4096]> encoder_layers_8_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(745153536)))];
            tensor<fp16, [1, 256, 4096]> linear_52_cast_fp16 = linear(bias = encoder_layers_8_fc1_bias_to_fp16, weight = encoder_layers_8_fc1_weight_to_fp16, x = input0_53_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_674_cast_fp16 = relu(x = linear_52_cast_fp16)[name = tensor<string, []>("op_674_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_8_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(745161792)))];
            tensor<fp16, [1024]> encoder_layers_8_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(753550464)))];
            tensor<fp16, [1, 256, 1024]> linear_53_cast_fp16 = linear(bias = encoder_layers_8_fc2_bias_to_fp16, weight = encoder_layers_8_fc2_weight_to_fp16, x = var_674_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_680_cast_fp16 = add(x = input_55_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("op_680_cast_fp16")];
            tensor<int32, [1]> hidden_states_39_axes_0 = const()[name = tensor<string, []>("hidden_states_39_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_9_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(753552576)))];
            tensor<fp16, [1024]> encoder_layers_9_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(753554688)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_39_cast_fp16 = layer_norm(axes = hidden_states_39_axes_0, beta = encoder_layers_9_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_9_self_attn_layer_norm_weight_to_fp16, x = var_680_cast_fp16)[name = tensor<string, []>("hidden_states_39_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(753556800)))];
            tensor<fp16, [1024]> encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(755654016)))];
            tensor<fp16, [1, 256, 1024]> linear_54_cast_fp16 = linear(bias = encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_q_proj_weight_to_fp16, x = hidden_states_39_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<int32, [4]> var_699 = const()[name = tensor<string, []>("op_699"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_700_cast_fp16 = reshape(shape = var_699, x = linear_54_cast_fp16)[name = tensor<string, []>("op_700_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(755656128)))];
            tensor<fp16, [1024]> encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757753344)))];
            tensor<fp16, [1, 256, 1024]> linear_55_cast_fp16 = linear(bias = encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_k_proj_weight_to_fp16, x = hidden_states_39_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757755456)))];
            tensor<fp16, [1024]> encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759852672)))];
            tensor<fp16, [1, 256, 1024]> linear_56_cast_fp16 = linear(bias = encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_v_proj_weight_to_fp16, x = hidden_states_39_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<int32, [4]> var_708 = const()[name = tensor<string, []>("op_708"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_709_cast_fp16 = reshape(shape = var_708, x = linear_55_cast_fp16)[name = tensor<string, []>("op_709_cast_fp16")];
            tensor<int32, [4]> var_711 = const()[name = tensor<string, []>("op_711"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_712_cast_fp16 = reshape(shape = var_711, x = linear_56_cast_fp16)[name = tensor<string, []>("op_712_cast_fp16")];
            tensor<int32, [4]> value_20_perm_0 = const()[name = tensor<string, []>("value_20_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_715_transpose_x_0 = const()[name = tensor<string, []>("op_715_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_715_transpose_y_0 = const()[name = tensor<string, []>("op_715_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_54_perm_0 = const()[name = tensor<string, []>("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_55_perm_0 = const()[name = tensor<string, []>("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_55 = transpose(perm = transpose_55_perm_0, x = var_709_cast_fp16)[name = tensor<string, []>("transpose_69")];
            tensor<fp16, [1, 16, 256, 64]> transpose_54 = transpose(perm = transpose_54_perm_0, x = var_700_cast_fp16)[name = tensor<string, []>("transpose_70")];
            tensor<fp16, [1, 16, 256, 256]> var_715_cast_fp16 = matmul(transpose_x = var_715_transpose_x_0, transpose_y = var_715_transpose_y_0, x = transpose_54, y = transpose_55)[name = tensor<string, []>("op_715_cast_fp16")];
            tensor<fp16, []> var_716_to_fp16 = const()[name = tensor<string, []>("op_716_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_20_cast_fp16 = mul(x = var_715_cast_fp16, y = var_716_to_fp16)[name = tensor<string, []>("attn_weights_20_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_59_cast_fp16 = add(x = attn_weights_20_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_57_cast_fp16 = softmax(axis = var_18, x = input_59_cast_fp16)[name = tensor<string, []>("input0_57_cast_fp16")];
            tensor<bool, []> attn_output_20_transpose_x_0 = const()[name = tensor<string, []>("attn_output_20_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_20_transpose_y_0 = const()[name = tensor<string, []>("attn_output_20_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_20_cast_fp16 = transpose(perm = value_20_perm_0, x = var_712_cast_fp16)[name = tensor<string, []>("transpose_71")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_20_cast_fp16 = matmul(transpose_x = attn_output_20_transpose_x_0, transpose_y = attn_output_20_transpose_y_0, x = input0_57_cast_fp16, y = value_20_cast_fp16)[name = tensor<string, []>("attn_output_20_cast_fp16")];
            tensor<int32, [4]> var_722_perm_0 = const()[name = tensor<string, []>("op_722_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_724 = const()[name = tensor<string, []>("op_724"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_722_cast_fp16 = transpose(perm = var_722_perm_0, x = attn_output_20_cast_fp16)[name = tensor<string, []>("transpose_68")];
            tensor<fp16, [1, 256, 1024]> var_725_cast_fp16 = reshape(shape = var_724, x = var_722_cast_fp16)[name = tensor<string, []>("op_725_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759854784)))];
            tensor<fp16, [1024]> encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(761952000)))];
            tensor<fp16, [1, 256, 1024]> linear_57_cast_fp16 = linear(bias = encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_out_proj_weight_to_fp16, x = var_725_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_61_cast_fp16 = add(x = var_680_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<int32, [1]> input0_59_axes_0 = const()[name = tensor<string, []>("input0_59_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_9_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(761954112)))];
            tensor<fp16, [1024]> encoder_layers_9_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(761956224)))];
            tensor<fp16, [1, 256, 1024]> input0_59_cast_fp16 = layer_norm(axes = input0_59_axes_0, beta = encoder_layers_9_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_9_final_layer_norm_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("input0_59_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_9_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(761958336)))];
            tensor<fp16, [4096]> encoder_layers_9_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(770347008)))];
            tensor<fp16, [1, 256, 4096]> linear_58_cast_fp16 = linear(bias = encoder_layers_9_fc1_bias_to_fp16, weight = encoder_layers_9_fc1_weight_to_fp16, x = input0_59_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_739_cast_fp16 = relu(x = linear_58_cast_fp16)[name = tensor<string, []>("op_739_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_9_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(770355264)))];
            tensor<fp16, [1024]> encoder_layers_9_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(778743936)))];
            tensor<fp16, [1, 256, 1024]> linear_59_cast_fp16 = linear(bias = encoder_layers_9_fc2_bias_to_fp16, weight = encoder_layers_9_fc2_weight_to_fp16, x = var_739_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_745_cast_fp16 = add(x = input_61_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("op_745_cast_fp16")];
            tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_10_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(778746048)))];
            tensor<fp16, [1024]> encoder_layers_10_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(778748160)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = encoder_layers_10_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_10_self_attn_layer_norm_weight_to_fp16, x = var_745_cast_fp16)[name = tensor<string, []>("hidden_states_43_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(778750272)))];
            tensor<fp16, [1024]> encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780847488)))];
            tensor<fp16, [1, 256, 1024]> linear_60_cast_fp16 = linear(bias = encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_q_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")];
            tensor<int32, [4]> var_764 = const()[name = tensor<string, []>("op_764"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_765_cast_fp16 = reshape(shape = var_764, x = linear_60_cast_fp16)[name = tensor<string, []>("op_765_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780849600)))];
            tensor<fp16, [1024]> encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782946816)))];
            tensor<fp16, [1, 256, 1024]> linear_61_cast_fp16 = linear(bias = encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_k_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782948928)))];
            tensor<fp16, [1024]> encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785046144)))];
            tensor<fp16, [1, 256, 1024]> linear_62_cast_fp16 = linear(bias = encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_v_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<int32, [4]> var_773 = const()[name = tensor<string, []>("op_773"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_774_cast_fp16 = reshape(shape = var_773, x = linear_61_cast_fp16)[name = tensor<string, []>("op_774_cast_fp16")];
            tensor<int32, [4]> var_776 = const()[name = tensor<string, []>("op_776"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_777_cast_fp16 = reshape(shape = var_776, x = linear_62_cast_fp16)[name = tensor<string, []>("op_777_cast_fp16")];
            tensor<int32, [4]> value_22_perm_0 = const()[name = tensor<string, []>("value_22_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_780_transpose_x_0 = const()[name = tensor<string, []>("op_780_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_780_transpose_y_0 = const()[name = tensor<string, []>("op_780_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_56_perm_0 = const()[name = tensor<string, []>("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_57_perm_0 = const()[name = tensor<string, []>("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_57 = transpose(perm = transpose_57_perm_0, x = var_774_cast_fp16)[name = tensor<string, []>("transpose_65")];
            tensor<fp16, [1, 16, 256, 64]> transpose_56 = transpose(perm = transpose_56_perm_0, x = var_765_cast_fp16)[name = tensor<string, []>("transpose_66")];
            tensor<fp16, [1, 16, 256, 256]> var_780_cast_fp16 = matmul(transpose_x = var_780_transpose_x_0, transpose_y = var_780_transpose_y_0, x = transpose_56, y = transpose_57)[name = tensor<string, []>("op_780_cast_fp16")];
            tensor<fp16, []> var_781_to_fp16 = const()[name = tensor<string, []>("op_781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_22_cast_fp16 = mul(x = var_780_cast_fp16, y = var_781_to_fp16)[name = tensor<string, []>("attn_weights_22_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_65_cast_fp16 = add(x = attn_weights_22_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_63_cast_fp16 = softmax(axis = var_18, x = input_65_cast_fp16)[name = tensor<string, []>("input0_63_cast_fp16")];
            tensor<bool, []> attn_output_22_transpose_x_0 = const()[name = tensor<string, []>("attn_output_22_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_22_transpose_y_0 = const()[name = tensor<string, []>("attn_output_22_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_22_cast_fp16 = transpose(perm = value_22_perm_0, x = var_777_cast_fp16)[name = tensor<string, []>("transpose_67")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_22_cast_fp16 = matmul(transpose_x = attn_output_22_transpose_x_0, transpose_y = attn_output_22_transpose_y_0, x = input0_63_cast_fp16, y = value_22_cast_fp16)[name = tensor<string, []>("attn_output_22_cast_fp16")];
            tensor<int32, [4]> var_787_perm_0 = const()[name = tensor<string, []>("op_787_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_789 = const()[name = tensor<string, []>("op_789"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_787_cast_fp16 = transpose(perm = var_787_perm_0, x = attn_output_22_cast_fp16)[name = tensor<string, []>("transpose_64")];
            tensor<fp16, [1, 256, 1024]> var_790_cast_fp16 = reshape(shape = var_789, x = var_787_cast_fp16)[name = tensor<string, []>("op_790_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785048256)))];
            tensor<fp16, [1024]> encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(787145472)))];
            tensor<fp16, [1, 256, 1024]> linear_63_cast_fp16 = linear(bias = encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_out_proj_weight_to_fp16, x = var_790_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_67_cast_fp16 = add(x = var_745_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")];
            tensor<int32, [1]> input0_65_axes_0 = const()[name = tensor<string, []>("input0_65_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_10_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(787147584)))];
            tensor<fp16, [1024]> encoder_layers_10_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(787149696)))];
            tensor<fp16, [1, 256, 1024]> input0_65_cast_fp16 = layer_norm(axes = input0_65_axes_0, beta = encoder_layers_10_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_10_final_layer_norm_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("input0_65_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_10_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(787151808)))];
            tensor<fp16, [4096]> encoder_layers_10_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795540480)))];
            tensor<fp16, [1, 256, 4096]> linear_64_cast_fp16 = linear(bias = encoder_layers_10_fc1_bias_to_fp16, weight = encoder_layers_10_fc1_weight_to_fp16, x = input0_65_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_804_cast_fp16 = relu(x = linear_64_cast_fp16)[name = tensor<string, []>("op_804_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_10_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795548736)))];
            tensor<fp16, [1024]> encoder_layers_10_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(803937408)))];
            tensor<fp16, [1, 256, 1024]> linear_65_cast_fp16 = linear(bias = encoder_layers_10_fc2_bias_to_fp16, weight = encoder_layers_10_fc2_weight_to_fp16, x = var_804_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_810_cast_fp16 = add(x = input_67_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("op_810_cast_fp16")];
            tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_11_self_attn_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(803939520)))];
            tensor<fp16, [1024]> encoder_layers_11_self_attn_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(803941632)))];
            tensor<fp16, [1, 256, 1024]> hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = encoder_layers_11_self_attn_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_11_self_attn_layer_norm_weight_to_fp16, x = var_810_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(803943744)))];
            tensor<fp16, [1024]> encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(806040960)))];
            tensor<fp16, [1, 256, 1024]> linear_66_cast_fp16 = linear(bias = encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_q_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<int32, [4]> var_829 = const()[name = tensor<string, []>("op_829"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_830_cast_fp16 = reshape(shape = var_829, x = linear_66_cast_fp16)[name = tensor<string, []>("op_830_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(806043072)))];
            tensor<fp16, [1024]> encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808140288)))];
            tensor<fp16, [1, 256, 1024]> linear_67_cast_fp16 = linear(bias = encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_k_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808142400)))];
            tensor<fp16, [1024]> encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810239616)))];
            tensor<fp16, [1, 256, 1024]> linear_68_cast_fp16 = linear(bias = encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_v_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<int32, [4]> var_838 = const()[name = tensor<string, []>("op_838"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_839_cast_fp16 = reshape(shape = var_838, x = linear_67_cast_fp16)[name = tensor<string, []>("op_839_cast_fp16")];
            tensor<int32, [4]> var_841 = const()[name = tensor<string, []>("op_841"), val = tensor<int32, [4]>([1, 256, -1, 64])];
            tensor<fp16, [1, 256, 16, 64]> var_842_cast_fp16 = reshape(shape = var_841, x = linear_68_cast_fp16)[name = tensor<string, []>("op_842_cast_fp16")];
            tensor<int32, [4]> value_1_perm_0 = const()[name = tensor<string, []>("value_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_845_transpose_x_0 = const()[name = tensor<string, []>("op_845_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_845_transpose_y_0 = const()[name = tensor<string, []>("op_845_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_58_perm_0 = const()[name = tensor<string, []>("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_59_perm_0 = const()[name = tensor<string, []>("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 64, 256]> transpose_59 = transpose(perm = transpose_59_perm_0, x = var_839_cast_fp16)[name = tensor<string, []>("transpose_61")];
            tensor<fp16, [1, 16, 256, 64]> transpose_58 = transpose(perm = transpose_58_perm_0, x = var_830_cast_fp16)[name = tensor<string, []>("transpose_62")];
            tensor<fp16, [1, 16, 256, 256]> var_845_cast_fp16 = matmul(transpose_x = var_845_transpose_x_0, transpose_y = var_845_transpose_y_0, x = transpose_58, y = transpose_59)[name = tensor<string, []>("op_845_cast_fp16")];
            tensor<fp16, []> var_846_to_fp16 = const()[name = tensor<string, []>("op_846_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 16, 256, 256]> attn_weights_1_cast_fp16 = mul(x = var_845_cast_fp16, y = var_846_to_fp16)[name = tensor<string, []>("attn_weights_1_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input_4_cast_fp16 = add(x = attn_weights_1_cast_fp16, y = attention_mask0_1_cast_fp16)[name = tensor<string, []>("input_4_cast_fp16")];
            tensor<fp16, [1, 16, 256, 256]> input0_4_cast_fp16 = softmax(axis = var_18, x = input_4_cast_fp16)[name = tensor<string, []>("input0_4_cast_fp16")];
            tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 256, 64]> value_1_cast_fp16 = transpose(perm = value_1_perm_0, x = var_842_cast_fp16)[name = tensor<string, []>("transpose_63")];
            tensor<fp16, [1, 16, 256, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input0_4_cast_fp16, y = value_1_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")];
            tensor<int32, [4]> var_852_perm_0 = const()[name = tensor<string, []>("op_852_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_854 = const()[name = tensor<string, []>("op_854"), val = tensor<int32, [3]>([1, 256, -1])];
            tensor<fp16, [1, 256, 16, 64]> var_852_cast_fp16 = transpose(perm = var_852_perm_0, x = attn_output_1_cast_fp16)[name = tensor<string, []>("transpose_60")];
            tensor<fp16, [1, 256, 1024]> var_855_cast_fp16 = reshape(shape = var_854, x = var_852_cast_fp16)[name = tensor<string, []>("op_855_cast_fp16")];
            tensor<fp16, [1024, 1024]> encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810241728)))];
            tensor<fp16, [1024]> encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812338944)))];
            tensor<fp16, [1, 256, 1024]> linear_69_cast_fp16 = linear(bias = encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_out_proj_weight_to_fp16, x = var_855_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> input_2_cast_fp16 = add(x = var_810_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("input_2_cast_fp16")];
            tensor<int32, [1]> input0_1_axes_0 = const()[name = tensor<string, []>("input0_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layers_11_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812341056)))];
            tensor<fp16, [1024]> encoder_layers_11_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812343168)))];
            tensor<fp16, [1, 256, 1024]> input0_1_cast_fp16 = layer_norm(axes = input0_1_axes_0, beta = encoder_layers_11_final_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layers_11_final_layer_norm_weight_to_fp16, x = input_2_cast_fp16)[name = tensor<string, []>("input0_1_cast_fp16")];
            tensor<fp16, [4096, 1024]> encoder_layers_11_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_fc1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812345280)))];
            tensor<fp16, [4096]> encoder_layers_11_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_fc1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(820733952)))];
            tensor<fp16, [1, 256, 4096]> linear_70_cast_fp16 = linear(bias = encoder_layers_11_fc1_bias_to_fp16, weight = encoder_layers_11_fc1_weight_to_fp16, x = input0_1_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
            tensor<fp16, [1, 256, 4096]> var_869_cast_fp16 = relu(x = linear_70_cast_fp16)[name = tensor<string, []>("op_869_cast_fp16")];
            tensor<fp16, [1024, 4096]> encoder_layers_11_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_fc2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(820742208)))];
            tensor<fp16, [1024]> encoder_layers_11_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_fc2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(829130880)))];
            tensor<fp16, [1, 256, 1024]> linear_71_cast_fp16 = linear(bias = encoder_layers_11_fc2_bias_to_fp16, weight = encoder_layers_11_fc2_weight_to_fp16, x = var_869_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
            tensor<fp16, [1, 256, 1024]> var_875_cast_fp16 = add(x = input_2_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("op_875_cast_fp16")];
            tensor<int32, [1]> var_879_axes_0 = const()[name = tensor<string, []>("op_879_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> encoder_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layer_norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(829132992)))];
            tensor<fp16, [1024]> encoder_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layer_norm_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(829135104)))];
            tensor<fp16, [1, 256, 1024]> var_879 = layer_norm(axes = var_879_axes_0, beta = encoder_layer_norm_bias_to_fp16, epsilon = var_6_to_fp16, gamma = encoder_layer_norm_weight_to_fp16, x = var_875_cast_fp16)[name = tensor<string, []>("op_879_cast_fp16")];
        } -> (var_879);
}