program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.12.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
    func main<ios17>(tensor<int32, [1, 64]> input_ids) {
            tensor<int32, []> inputs_embeds_batch_dims_0 = const()[name = tensor<string, []>("inputs_embeds_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<bool, []> inputs_embeds_validate_indices_0 = const()[name = tensor<string, []>("inputs_embeds_validate_indices_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [32000, 1152]> model_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor<fp16, [32000, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<string, []> input_ids_to_int16_dtype_0 = const()[name = tensor<string, []>("input_ids_to_int16_dtype_0"), val = tensor<string, []>("int16")];
            tensor<string, []> cast_114_dtype_0 = const()[name = tensor<string, []>("cast_114_dtype_0"), val = tensor<string, []>("int32")];
            tensor<int32, []> greater_equal_0_y_0 = const()[name = tensor<string, []>("greater_equal_0_y_0"), val = tensor<int32, []>(0)];
            tensor<int16, [1, 64]> input_ids_to_int16 = cast(dtype = input_ids_to_int16_dtype_0, x = input_ids)[name = tensor<string, []>("cast_117")];
            tensor<int32, [1, 64]> cast_114 = cast(dtype = cast_114_dtype_0, x = input_ids_to_int16)[name = tensor<string, []>("cast_116")];
            tensor<bool, [1, 64]> greater_equal_0 = greater_equal(x = cast_114, 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, []>(32000)];
            tensor<int32, [1, 64]> add_0 = add(x = cast_114, y = slice_by_index_0)[name = tensor<string, []>("add_0")];
            tensor<int32, [1, 64]> select_0 = select(a = cast_114, b = add_0, cond = greater_equal_0)[name = tensor<string, []>("select_0")];
            tensor<int32, []> inputs_embeds_cast_fp16_cast_uint16_axis_0 = const()[name = tensor<string, []>("inputs_embeds_cast_fp16_cast_uint16_axis_0"), val = tensor<int32, []>(0)];
            tensor<string, []> select_0_to_int16_dtype_0 = const()[name = tensor<string, []>("select_0_to_int16_dtype_0"), val = tensor<string, []>("int16")];
            tensor<int16, [1, 64]> select_0_to_int16 = cast(dtype = select_0_to_int16_dtype_0, x = select_0)[name = tensor<string, []>("cast_115")];
            tensor<fp16, [1, 64, 1152]> inputs_embeds_cast_fp16_cast_uint16_cast_uint16 = gather(axis = inputs_embeds_cast_fp16_cast_uint16_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = select_0_to_int16, validate_indices = inputs_embeds_validate_indices_0, x = model_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor<string, []>("inputs_embeds_cast_fp16_cast_uint16_cast_uint16")];
            tensor<fp16, [1, 64, 1152]> position_embeddings_to_fp16 = const()[name = tensor<string, []>("position_embeddings_to_fp16"), val = tensor<fp16, [1, 64, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73728128)))];
            tensor<fp16, [1, 64, 1152]> input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16_cast_uint16_cast_uint16, y = position_embeddings_to_fp16)[name = tensor<string, []>("input_3_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, [1152]> model_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73875648)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73878016)))];
            tensor<fp16, []> var_11_to_fp16 = const()[name = tensor<string, []>("op_11_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 64, 1152]> hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = model_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73880384)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76534656)))];
            tensor<fp16, [1, 64, 1152]> linear_0_cast_fp16 = linear(bias = model_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76537024)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79191296)))];
            tensor<fp16, [1, 64, 1152]> linear_1_cast_fp16 = linear(bias = model_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79193664)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81847936)))];
            tensor<fp16, [1, 64, 1152]> linear_2_cast_fp16 = linear(bias = model_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> var_113 = const()[name = tensor<string, []>("op_113"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_114_cast_fp16 = reshape(shape = var_113, x = linear_0_cast_fp16)[name = tensor<string, []>("op_114_cast_fp16")];
            tensor<int32, [4]> var_116 = const()[name = tensor<string, []>("op_116"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_117_cast_fp16 = reshape(shape = var_116, x = linear_1_cast_fp16)[name = tensor<string, []>("op_117_cast_fp16")];
            tensor<int32, [4]> var_119 = const()[name = tensor<string, []>("op_119"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_120_cast_fp16 = reshape(shape = var_119, x = linear_2_cast_fp16)[name = tensor<string, []>("op_120_cast_fp16")];
            tensor<int32, [4]> value_states_3_perm_0 = const()[name = tensor<string, []>("value_states_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_0_y_0_to_fp16 = const()[name = tensor<string, []>("mul_0_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_0_cast_fp16 = mul(x = var_114_cast_fp16, y = mul_0_y_0_to_fp16)[name = tensor<string, []>("mul_0_cast_fp16")];
            tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_108_perm_0 = const()[name = tensor<string, []>("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_109_perm_0 = const()[name = tensor<string, []>("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_109 = transpose(perm = transpose_109_perm_0, x = var_117_cast_fp16)[name = tensor<string, []>("transpose_267")];
            tensor<fp16, [1, 16, 64, 72]> transpose_108 = transpose(perm = transpose_108_perm_0, x = mul_0_cast_fp16)[name = tensor<string, []>("transpose_268")];
            tensor<fp16, [1, 16, 64, 64]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_108, y = transpose_109)[name = tensor<string, []>("matmul_0_cast_fp16")];
            tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = matmul_0_cast_fp16)[name = tensor<string, []>("softmax_0_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, 64, 72]> value_states_3_cast_fp16 = transpose(perm = value_states_3_perm_0, x = var_120_cast_fp16)[name = tensor<string, []>("transpose_269")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0_cast_fp16, y = value_states_3_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")];
            tensor<int32, [4]> var_123_perm_0 = const()[name = tensor<string, []>("op_123_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_125 = const()[name = tensor<string, []>("op_125"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_123_cast_fp16 = transpose(perm = var_123_perm_0, x = attn_output_1_cast_fp16)[name = tensor<string, []>("transpose_266")];
            tensor<fp16, [1, 64, 1152]> input_5_cast_fp16 = reshape(shape = var_125, x = var_123_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81850304)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84504576)))];
            tensor<fp16, [1, 64, 1152]> linear_3_cast_fp16 = linear(bias = model_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_7_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<int32, [1]> input_9_axes_0 = const()[name = tensor<string, []>("input_9_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84506944)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84509312)))];
            tensor<fp16, [1, 64, 1152]> input_9_cast_fp16 = layer_norm(axes = input_9_axes_0, beta = model_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84511680)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94428160)))];
            tensor<fp16, [1, 64, 4304]> linear_4_cast_fp16 = linear(bias = model_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<string, []> input_13_mode_0 = const()[name = tensor<string, []>("input_13_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_13_cast_fp16 = gelu(mode = input_13_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94436864)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104353344)))];
            tensor<fp16, [1, 64, 1152]> linear_5_cast_fp16 = linear(bias = model_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_15_cast_fp16 = add(x = input_7_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_15_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, [1152]> model_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104355712)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104358080)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = model_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104360448)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107014720)))];
            tensor<fp16, [1, 64, 1152]> linear_6_cast_fp16 = linear(bias = model_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107017088)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109671360)))];
            tensor<fp16, [1, 64, 1152]> linear_7_cast_fp16 = linear(bias = model_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = model_text_model_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, [1152, 1152]> model_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109673728)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112328000)))];
            tensor<fp16, [1, 64, 1152]> linear_8_cast_fp16 = linear(bias = model_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = model_text_model_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_168 = const()[name = tensor<string, []>("op_168"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_169_cast_fp16 = reshape(shape = var_168, x = linear_6_cast_fp16)[name = tensor<string, []>("op_169_cast_fp16")];
            tensor<int32, [4]> var_171 = const()[name = tensor<string, []>("op_171"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_172_cast_fp16 = reshape(shape = var_171, x = linear_7_cast_fp16)[name = tensor<string, []>("op_172_cast_fp16")];
            tensor<int32, [4]> var_174 = const()[name = tensor<string, []>("op_174"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_175_cast_fp16 = reshape(shape = var_174, x = linear_8_cast_fp16)[name = tensor<string, []>("op_175_cast_fp16")];
            tensor<int32, [4]> value_states_7_perm_0 = const()[name = tensor<string, []>("value_states_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_1_y_0_to_fp16 = const()[name = tensor<string, []>("mul_1_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_1_cast_fp16 = mul(x = var_169_cast_fp16, y = mul_1_y_0_to_fp16)[name = tensor<string, []>("mul_1_cast_fp16")];
            tensor<bool, []> matmul_1_transpose_y_0 = const()[name = tensor<string, []>("matmul_1_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_1_transpose_x_0 = const()[name = tensor<string, []>("matmul_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_110_perm_0 = const()[name = tensor<string, []>("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_111_perm_0 = const()[name = tensor<string, []>("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_111 = transpose(perm = transpose_111_perm_0, x = var_172_cast_fp16)[name = tensor<string, []>("transpose_263")];
            tensor<fp16, [1, 16, 64, 72]> transpose_110 = transpose(perm = transpose_110_perm_0, x = mul_1_cast_fp16)[name = tensor<string, []>("transpose_264")];
            tensor<fp16, [1, 16, 64, 64]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = transpose_110, y = transpose_111)[name = tensor<string, []>("matmul_1_cast_fp16")];
            tensor<int32, []> softmax_1_axis_0 = const()[name = tensor<string, []>("softmax_1_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = matmul_1_cast_fp16)[name = tensor<string, []>("softmax_1_cast_fp16")];
            tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_7_cast_fp16 = transpose(perm = value_states_7_perm_0, x = var_175_cast_fp16)[name = tensor<string, []>("transpose_265")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = softmax_1_cast_fp16, y = value_states_7_cast_fp16)[name = tensor<string, []>("attn_output_5_cast_fp16")];
            tensor<int32, [4]> var_178_perm_0 = const()[name = tensor<string, []>("op_178_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_180 = const()[name = tensor<string, []>("op_180"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_178_cast_fp16 = transpose(perm = var_178_perm_0, x = attn_output_5_cast_fp16)[name = tensor<string, []>("transpose_262")];
            tensor<fp16, [1, 64, 1152]> input_17_cast_fp16 = reshape(shape = var_180, x = var_178_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112330368)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114984640)))];
            tensor<fp16, [1, 64, 1152]> linear_9_cast_fp16 = linear(bias = model_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_19_cast_fp16 = add(x = input_15_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
            tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114987008)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114989376)))];
            tensor<fp16, [1, 64, 1152]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = model_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114991744)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124908224)))];
            tensor<fp16, [1, 64, 4304]> linear_10_cast_fp16 = linear(bias = model_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
            tensor<string, []> input_25_mode_0 = const()[name = tensor<string, []>("input_25_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_25_cast_fp16 = gelu(mode = input_25_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124916928)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134833408)))];
            tensor<fp16, [1, 64, 1152]> linear_11_cast_fp16 = linear(bias = model_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")];
            tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = tensor<string, []>("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134835776)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134838144)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = model_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134840512)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137494784)))];
            tensor<fp16, [1, 64, 1152]> linear_12_cast_fp16 = linear(bias = model_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137497152)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140151424)))];
            tensor<fp16, [1, 64, 1152]> linear_13_cast_fp16 = linear(bias = model_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140153792)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142808064)))];
            tensor<fp16, [1, 64, 1152]> linear_14_cast_fp16 = linear(bias = model_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<int32, [4]> var_223 = const()[name = tensor<string, []>("op_223"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_224_cast_fp16 = reshape(shape = var_223, x = linear_12_cast_fp16)[name = tensor<string, []>("op_224_cast_fp16")];
            tensor<int32, [4]> var_226 = const()[name = tensor<string, []>("op_226"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_227_cast_fp16 = reshape(shape = var_226, x = linear_13_cast_fp16)[name = tensor<string, []>("op_227_cast_fp16")];
            tensor<int32, [4]> var_229 = const()[name = tensor<string, []>("op_229"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_230_cast_fp16 = reshape(shape = var_229, x = linear_14_cast_fp16)[name = tensor<string, []>("op_230_cast_fp16")];
            tensor<int32, [4]> value_states_11_perm_0 = const()[name = tensor<string, []>("value_states_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_2_y_0_to_fp16 = const()[name = tensor<string, []>("mul_2_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_2_cast_fp16 = mul(x = var_224_cast_fp16, y = mul_2_y_0_to_fp16)[name = tensor<string, []>("mul_2_cast_fp16")];
            tensor<bool, []> matmul_2_transpose_y_0 = const()[name = tensor<string, []>("matmul_2_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_2_transpose_x_0 = const()[name = tensor<string, []>("matmul_2_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_112_perm_0 = const()[name = tensor<string, []>("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_113_perm_0 = const()[name = tensor<string, []>("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_113 = transpose(perm = transpose_113_perm_0, x = var_227_cast_fp16)[name = tensor<string, []>("transpose_259")];
            tensor<fp16, [1, 16, 64, 72]> transpose_112 = transpose(perm = transpose_112_perm_0, x = mul_2_cast_fp16)[name = tensor<string, []>("transpose_260")];
            tensor<fp16, [1, 16, 64, 64]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_112, y = transpose_113)[name = tensor<string, []>("matmul_2_cast_fp16")];
            tensor<int32, []> softmax_2_axis_0 = const()[name = tensor<string, []>("softmax_2_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_2_cast_fp16 = softmax(axis = softmax_2_axis_0, x = matmul_2_cast_fp16)[name = tensor<string, []>("softmax_2_cast_fp16")];
            tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_11_cast_fp16 = transpose(perm = value_states_11_perm_0, x = var_230_cast_fp16)[name = tensor<string, []>("transpose_261")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = softmax_2_cast_fp16, y = value_states_11_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")];
            tensor<int32, [4]> var_233_perm_0 = const()[name = tensor<string, []>("op_233_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_235 = const()[name = tensor<string, []>("op_235"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_233_cast_fp16 = transpose(perm = var_233_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_258")];
            tensor<fp16, [1, 64, 1152]> input_29_cast_fp16 = reshape(shape = var_235, x = var_233_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142810432)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145464704)))];
            tensor<fp16, [1, 64, 1152]> linear_15_cast_fp16 = linear(bias = model_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16, x = input_29_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_31_cast_fp16 = add(x = input_27_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
            tensor<int32, [1]> input_33_axes_0 = const()[name = tensor<string, []>("input_33_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145467072)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145469440)))];
            tensor<fp16, [1, 64, 1152]> input_33_cast_fp16 = layer_norm(axes = input_33_axes_0, beta = model_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145471808)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155388288)))];
            tensor<fp16, [1, 64, 4304]> linear_16_cast_fp16 = linear(bias = model_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
            tensor<string, []> input_37_mode_0 = const()[name = tensor<string, []>("input_37_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_37_cast_fp16 = gelu(mode = input_37_mode_0, x = linear_16_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155396992)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165313472)))];
            tensor<fp16, [1, 64, 1152]> linear_17_cast_fp16 = linear(bias = model_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16, x = input_37_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("input_39_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, [1152]> model_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165315840)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165318208)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = model_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165320576)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167974848)))];
            tensor<fp16, [1, 64, 1152]> linear_18_cast_fp16 = linear(bias = model_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167977216)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170631488)))];
            tensor<fp16, [1, 64, 1152]> linear_19_cast_fp16 = linear(bias = model_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170633856)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173288128)))];
            tensor<fp16, [1, 64, 1152]> linear_20_cast_fp16 = linear(bias = model_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<int32, [4]> var_278 = const()[name = tensor<string, []>("op_278"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_279_cast_fp16 = reshape(shape = var_278, x = linear_18_cast_fp16)[name = tensor<string, []>("op_279_cast_fp16")];
            tensor<int32, [4]> var_281 = const()[name = tensor<string, []>("op_281"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_282_cast_fp16 = reshape(shape = var_281, x = linear_19_cast_fp16)[name = tensor<string, []>("op_282_cast_fp16")];
            tensor<int32, [4]> var_284 = const()[name = tensor<string, []>("op_284"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_285_cast_fp16 = reshape(shape = var_284, x = linear_20_cast_fp16)[name = tensor<string, []>("op_285_cast_fp16")];
            tensor<int32, [4]> value_states_15_perm_0 = const()[name = tensor<string, []>("value_states_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_3_y_0_to_fp16 = const()[name = tensor<string, []>("mul_3_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_3_cast_fp16 = mul(x = var_279_cast_fp16, y = mul_3_y_0_to_fp16)[name = tensor<string, []>("mul_3_cast_fp16")];
            tensor<bool, []> matmul_3_transpose_y_0 = const()[name = tensor<string, []>("matmul_3_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_3_transpose_x_0 = const()[name = tensor<string, []>("matmul_3_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_114_perm_0 = const()[name = tensor<string, []>("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_115_perm_0 = const()[name = tensor<string, []>("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_115 = transpose(perm = transpose_115_perm_0, x = var_282_cast_fp16)[name = tensor<string, []>("transpose_255")];
            tensor<fp16, [1, 16, 64, 72]> transpose_114 = transpose(perm = transpose_114_perm_0, x = mul_3_cast_fp16)[name = tensor<string, []>("transpose_256")];
            tensor<fp16, [1, 16, 64, 64]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = transpose_114, y = transpose_115)[name = tensor<string, []>("matmul_3_cast_fp16")];
            tensor<int32, []> softmax_3_axis_0 = const()[name = tensor<string, []>("softmax_3_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_3_cast_fp16 = softmax(axis = softmax_3_axis_0, x = matmul_3_cast_fp16)[name = tensor<string, []>("softmax_3_cast_fp16")];
            tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_15_cast_fp16 = transpose(perm = value_states_15_perm_0, x = var_285_cast_fp16)[name = tensor<string, []>("transpose_257")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = softmax_3_cast_fp16, y = value_states_15_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")];
            tensor<int32, [4]> var_288_perm_0 = const()[name = tensor<string, []>("op_288_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_290 = const()[name = tensor<string, []>("op_290"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_288_cast_fp16 = transpose(perm = var_288_perm_0, x = attn_output_13_cast_fp16)[name = tensor<string, []>("transpose_254")];
            tensor<fp16, [1, 64, 1152]> input_41_cast_fp16 = reshape(shape = var_290, x = var_288_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173290496)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175944768)))];
            tensor<fp16, [1, 64, 1152]> linear_21_cast_fp16 = linear(bias = model_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_43_cast_fp16 = add(x = input_39_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")];
            tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175947136)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175949504)))];
            tensor<fp16, [1, 64, 1152]> input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = model_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175951872)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185868352)))];
            tensor<fp16, [1, 64, 4304]> linear_22_cast_fp16 = linear(bias = model_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
            tensor<string, []> input_49_mode_0 = const()[name = tensor<string, []>("input_49_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = linear_22_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185877056)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195793536)))];
            tensor<fp16, [1, 64, 1152]> linear_23_cast_fp16 = linear(bias = model_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")];
            tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195795904)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195798272)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = model_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("hidden_states_25_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195800640)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198454912)))];
            tensor<fp16, [1, 64, 1152]> linear_24_cast_fp16 = linear(bias = model_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198457280)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201111552)))];
            tensor<fp16, [1, 64, 1152]> linear_25_cast_fp16 = linear(bias = model_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201113920)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203768192)))];
            tensor<fp16, [1, 64, 1152]> linear_26_cast_fp16 = linear(bias = model_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<int32, [4]> var_333 = const()[name = tensor<string, []>("op_333"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_334_cast_fp16 = reshape(shape = var_333, x = linear_24_cast_fp16)[name = tensor<string, []>("op_334_cast_fp16")];
            tensor<int32, [4]> var_336 = const()[name = tensor<string, []>("op_336"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_337_cast_fp16 = reshape(shape = var_336, x = linear_25_cast_fp16)[name = tensor<string, []>("op_337_cast_fp16")];
            tensor<int32, [4]> var_339 = const()[name = tensor<string, []>("op_339"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_340_cast_fp16 = reshape(shape = var_339, x = linear_26_cast_fp16)[name = tensor<string, []>("op_340_cast_fp16")];
            tensor<int32, [4]> value_states_19_perm_0 = const()[name = tensor<string, []>("value_states_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_4_y_0_to_fp16 = const()[name = tensor<string, []>("mul_4_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_4_cast_fp16 = mul(x = var_334_cast_fp16, y = mul_4_y_0_to_fp16)[name = tensor<string, []>("mul_4_cast_fp16")];
            tensor<bool, []> matmul_4_transpose_y_0 = const()[name = tensor<string, []>("matmul_4_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_4_transpose_x_0 = const()[name = tensor<string, []>("matmul_4_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_116_perm_0 = const()[name = tensor<string, []>("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_117_perm_0 = const()[name = tensor<string, []>("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_117 = transpose(perm = transpose_117_perm_0, x = var_337_cast_fp16)[name = tensor<string, []>("transpose_251")];
            tensor<fp16, [1, 16, 64, 72]> transpose_116 = transpose(perm = transpose_116_perm_0, x = mul_4_cast_fp16)[name = tensor<string, []>("transpose_252")];
            tensor<fp16, [1, 16, 64, 64]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_116, y = transpose_117)[name = tensor<string, []>("matmul_4_cast_fp16")];
            tensor<int32, []> softmax_4_axis_0 = const()[name = tensor<string, []>("softmax_4_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_4_cast_fp16 = softmax(axis = softmax_4_axis_0, x = matmul_4_cast_fp16)[name = tensor<string, []>("softmax_4_cast_fp16")];
            tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_19_cast_fp16 = transpose(perm = value_states_19_perm_0, x = var_340_cast_fp16)[name = tensor<string, []>("transpose_253")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_17_cast_fp16 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = softmax_4_cast_fp16, y = value_states_19_cast_fp16)[name = tensor<string, []>("attn_output_17_cast_fp16")];
            tensor<int32, [4]> var_343_perm_0 = const()[name = tensor<string, []>("op_343_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_345 = const()[name = tensor<string, []>("op_345"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_343_cast_fp16 = transpose(perm = var_343_perm_0, x = attn_output_17_cast_fp16)[name = tensor<string, []>("transpose_250")];
            tensor<fp16, [1, 64, 1152]> input_53_cast_fp16 = reshape(shape = var_345, x = var_343_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203770560)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206424832)))];
            tensor<fp16, [1, 64, 1152]> linear_27_cast_fp16 = linear(bias = model_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16, x = input_53_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_55_cast_fp16 = add(x = input_51_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
            tensor<int32, [1]> input_57_axes_0 = const()[name = tensor<string, []>("input_57_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206427200)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206429568)))];
            tensor<fp16, [1, 64, 1152]> input_57_cast_fp16 = layer_norm(axes = input_57_axes_0, beta = model_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206431936)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216348416)))];
            tensor<fp16, [1, 64, 4304]> linear_28_cast_fp16 = linear(bias = model_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
            tensor<string, []> input_61_mode_0 = const()[name = tensor<string, []>("input_61_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_61_cast_fp16 = gelu(mode = input_61_mode_0, x = linear_28_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216357120)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226273600)))];
            tensor<fp16, [1, 64, 1152]> linear_29_cast_fp16 = linear(bias = model_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_63_cast_fp16 = add(x = input_55_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("input_63_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, [1152]> model_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226275968)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226278336)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = model_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("hidden_states_31_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226280704)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228934976)))];
            tensor<fp16, [1, 64, 1152]> linear_30_cast_fp16 = linear(bias = model_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(228937344)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231591616)))];
            tensor<fp16, [1, 64, 1152]> linear_31_cast_fp16 = linear(bias = model_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231593984)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234248256)))];
            tensor<fp16, [1, 64, 1152]> linear_32_cast_fp16 = linear(bias = model_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<int32, [4]> var_388 = const()[name = tensor<string, []>("op_388"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_389_cast_fp16 = reshape(shape = var_388, x = linear_30_cast_fp16)[name = tensor<string, []>("op_389_cast_fp16")];
            tensor<int32, [4]> var_391 = const()[name = tensor<string, []>("op_391"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_392_cast_fp16 = reshape(shape = var_391, x = linear_31_cast_fp16)[name = tensor<string, []>("op_392_cast_fp16")];
            tensor<int32, [4]> var_394 = const()[name = tensor<string, []>("op_394"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_395_cast_fp16 = reshape(shape = var_394, x = linear_32_cast_fp16)[name = tensor<string, []>("op_395_cast_fp16")];
            tensor<int32, [4]> value_states_23_perm_0 = const()[name = tensor<string, []>("value_states_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_5_y_0_to_fp16 = const()[name = tensor<string, []>("mul_5_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_5_cast_fp16 = mul(x = var_389_cast_fp16, y = mul_5_y_0_to_fp16)[name = tensor<string, []>("mul_5_cast_fp16")];
            tensor<bool, []> matmul_5_transpose_y_0 = const()[name = tensor<string, []>("matmul_5_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_5_transpose_x_0 = const()[name = tensor<string, []>("matmul_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_118_perm_0 = const()[name = tensor<string, []>("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_119_perm_0 = const()[name = tensor<string, []>("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_119 = transpose(perm = transpose_119_perm_0, x = var_392_cast_fp16)[name = tensor<string, []>("transpose_247")];
            tensor<fp16, [1, 16, 64, 72]> transpose_118 = transpose(perm = transpose_118_perm_0, x = mul_5_cast_fp16)[name = tensor<string, []>("transpose_248")];
            tensor<fp16, [1, 16, 64, 64]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = transpose_118, y = transpose_119)[name = tensor<string, []>("matmul_5_cast_fp16")];
            tensor<int32, []> softmax_5_axis_0 = const()[name = tensor<string, []>("softmax_5_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_5_cast_fp16 = softmax(axis = softmax_5_axis_0, x = matmul_5_cast_fp16)[name = tensor<string, []>("softmax_5_cast_fp16")];
            tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_23_cast_fp16 = transpose(perm = value_states_23_perm_0, x = var_395_cast_fp16)[name = tensor<string, []>("transpose_249")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_21_cast_fp16 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = softmax_5_cast_fp16, y = value_states_23_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")];
            tensor<int32, [4]> var_398_perm_0 = const()[name = tensor<string, []>("op_398_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_400 = const()[name = tensor<string, []>("op_400"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_398_cast_fp16 = transpose(perm = var_398_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_246")];
            tensor<fp16, [1, 64, 1152]> input_65_cast_fp16 = reshape(shape = var_400, x = var_398_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234250624)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236904896)))];
            tensor<fp16, [1, 64, 1152]> linear_33_cast_fp16 = linear(bias = model_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_67_cast_fp16 = add(x = input_63_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")];
            tensor<int32, [1]> input_69_axes_0 = const()[name = tensor<string, []>("input_69_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236907264)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236909632)))];
            tensor<fp16, [1, 64, 1152]> input_69_cast_fp16 = layer_norm(axes = input_69_axes_0, beta = model_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236912000)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246828480)))];
            tensor<fp16, [1, 64, 4304]> linear_34_cast_fp16 = linear(bias = model_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16, x = input_69_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
            tensor<string, []> input_73_mode_0 = const()[name = tensor<string, []>("input_73_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_73_cast_fp16 = gelu(mode = input_73_mode_0, x = linear_34_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246837184)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256753664)))];
            tensor<fp16, [1, 64, 1152]> linear_35_cast_fp16 = linear(bias = model_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")];
            tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = tensor<string, []>("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256756032)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256758400)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = model_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("hidden_states_37_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256760768)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259415040)))];
            tensor<fp16, [1, 64, 1152]> linear_36_cast_fp16 = linear(bias = model_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259417408)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262071680)))];
            tensor<fp16, [1, 64, 1152]> linear_37_cast_fp16 = linear(bias = model_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262074048)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264728320)))];
            tensor<fp16, [1, 64, 1152]> linear_38_cast_fp16 = linear(bias = model_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<int32, [4]> var_443 = const()[name = tensor<string, []>("op_443"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_444_cast_fp16 = reshape(shape = var_443, x = linear_36_cast_fp16)[name = tensor<string, []>("op_444_cast_fp16")];
            tensor<int32, [4]> var_446 = const()[name = tensor<string, []>("op_446"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_447_cast_fp16 = reshape(shape = var_446, x = linear_37_cast_fp16)[name = tensor<string, []>("op_447_cast_fp16")];
            tensor<int32, [4]> var_449 = const()[name = tensor<string, []>("op_449"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_450_cast_fp16 = reshape(shape = var_449, x = linear_38_cast_fp16)[name = tensor<string, []>("op_450_cast_fp16")];
            tensor<int32, [4]> value_states_27_perm_0 = const()[name = tensor<string, []>("value_states_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_6_y_0_to_fp16 = const()[name = tensor<string, []>("mul_6_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_6_cast_fp16 = mul(x = var_444_cast_fp16, y = mul_6_y_0_to_fp16)[name = tensor<string, []>("mul_6_cast_fp16")];
            tensor<bool, []> matmul_6_transpose_y_0 = const()[name = tensor<string, []>("matmul_6_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_6_transpose_x_0 = const()[name = tensor<string, []>("matmul_6_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_120_perm_0 = const()[name = tensor<string, []>("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_121_perm_0 = const()[name = tensor<string, []>("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_121 = transpose(perm = transpose_121_perm_0, x = var_447_cast_fp16)[name = tensor<string, []>("transpose_243")];
            tensor<fp16, [1, 16, 64, 72]> transpose_120 = transpose(perm = transpose_120_perm_0, x = mul_6_cast_fp16)[name = tensor<string, []>("transpose_244")];
            tensor<fp16, [1, 16, 64, 64]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = transpose_120, y = transpose_121)[name = tensor<string, []>("matmul_6_cast_fp16")];
            tensor<int32, []> softmax_6_axis_0 = const()[name = tensor<string, []>("softmax_6_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_6_cast_fp16 = softmax(axis = softmax_6_axis_0, x = matmul_6_cast_fp16)[name = tensor<string, []>("softmax_6_cast_fp16")];
            tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_27_cast_fp16 = transpose(perm = value_states_27_perm_0, x = var_450_cast_fp16)[name = tensor<string, []>("transpose_245")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = softmax_6_cast_fp16, y = value_states_27_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")];
            tensor<int32, [4]> var_453_perm_0 = const()[name = tensor<string, []>("op_453_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_455 = const()[name = tensor<string, []>("op_455"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_453_cast_fp16 = transpose(perm = var_453_perm_0, x = attn_output_25_cast_fp16)[name = tensor<string, []>("transpose_242")];
            tensor<fp16, [1, 64, 1152]> input_77_cast_fp16 = reshape(shape = var_455, x = var_453_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264730688)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(267384960)))];
            tensor<fp16, [1, 64, 1152]> linear_39_cast_fp16 = linear(bias = model_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16, x = input_77_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_79_cast_fp16 = add(x = input_75_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")];
            tensor<int32, [1]> input_81_axes_0 = const()[name = tensor<string, []>("input_81_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(267387328)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(267389696)))];
            tensor<fp16, [1, 64, 1152]> input_81_cast_fp16 = layer_norm(axes = input_81_axes_0, beta = model_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_79_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(267392064)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277308544)))];
            tensor<fp16, [1, 64, 4304]> linear_40_cast_fp16 = linear(bias = model_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
            tensor<string, []> input_85_mode_0 = const()[name = tensor<string, []>("input_85_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_85_cast_fp16 = gelu(mode = input_85_mode_0, x = linear_40_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277317248)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287233728)))];
            tensor<fp16, [1, 64, 1152]> linear_41_cast_fp16 = linear(bias = model_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16, x = input_85_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_87_cast_fp16 = add(x = input_79_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("input_87_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, [1152]> model_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287236096)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287238464)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = model_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_87_cast_fp16)[name = tensor<string, []>("hidden_states_43_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287240832)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289895104)))];
            tensor<fp16, [1, 64, 1152]> linear_42_cast_fp16 = linear(bias = model_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289897472)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292551744)))];
            tensor<fp16, [1, 64, 1152]> linear_43_cast_fp16 = linear(bias = model_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292554112)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295208384)))];
            tensor<fp16, [1, 64, 1152]> linear_44_cast_fp16 = linear(bias = model_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<int32, [4]> var_498 = const()[name = tensor<string, []>("op_498"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_499_cast_fp16 = reshape(shape = var_498, x = linear_42_cast_fp16)[name = tensor<string, []>("op_499_cast_fp16")];
            tensor<int32, [4]> var_501 = const()[name = tensor<string, []>("op_501"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_502_cast_fp16 = reshape(shape = var_501, x = linear_43_cast_fp16)[name = tensor<string, []>("op_502_cast_fp16")];
            tensor<int32, [4]> var_504 = const()[name = tensor<string, []>("op_504"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_505_cast_fp16 = reshape(shape = var_504, x = linear_44_cast_fp16)[name = tensor<string, []>("op_505_cast_fp16")];
            tensor<int32, [4]> value_states_31_perm_0 = const()[name = tensor<string, []>("value_states_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_7_y_0_to_fp16 = const()[name = tensor<string, []>("mul_7_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_7_cast_fp16 = mul(x = var_499_cast_fp16, y = mul_7_y_0_to_fp16)[name = tensor<string, []>("mul_7_cast_fp16")];
            tensor<bool, []> matmul_7_transpose_y_0 = const()[name = tensor<string, []>("matmul_7_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_7_transpose_x_0 = const()[name = tensor<string, []>("matmul_7_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_122_perm_0 = const()[name = tensor<string, []>("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_123_perm_0 = const()[name = tensor<string, []>("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_123 = transpose(perm = transpose_123_perm_0, x = var_502_cast_fp16)[name = tensor<string, []>("transpose_239")];
            tensor<fp16, [1, 16, 64, 72]> transpose_122 = transpose(perm = transpose_122_perm_0, x = mul_7_cast_fp16)[name = tensor<string, []>("transpose_240")];
            tensor<fp16, [1, 16, 64, 64]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = transpose_122, y = transpose_123)[name = tensor<string, []>("matmul_7_cast_fp16")];
            tensor<int32, []> softmax_7_axis_0 = const()[name = tensor<string, []>("softmax_7_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_7_cast_fp16 = softmax(axis = softmax_7_axis_0, x = matmul_7_cast_fp16)[name = tensor<string, []>("softmax_7_cast_fp16")];
            tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_31_cast_fp16 = transpose(perm = value_states_31_perm_0, x = var_505_cast_fp16)[name = tensor<string, []>("transpose_241")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_29_cast_fp16 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = softmax_7_cast_fp16, y = value_states_31_cast_fp16)[name = tensor<string, []>("attn_output_29_cast_fp16")];
            tensor<int32, [4]> var_508_perm_0 = const()[name = tensor<string, []>("op_508_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_510 = const()[name = tensor<string, []>("op_510"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_508_cast_fp16 = transpose(perm = var_508_perm_0, x = attn_output_29_cast_fp16)[name = tensor<string, []>("transpose_238")];
            tensor<fp16, [1, 64, 1152]> input_89_cast_fp16 = reshape(shape = var_510, x = var_508_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295210752)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297865024)))];
            tensor<fp16, [1, 64, 1152]> linear_45_cast_fp16 = linear(bias = model_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16, x = input_89_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_91_cast_fp16 = add(x = input_87_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")];
            tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297867392)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297869760)))];
            tensor<fp16, [1, 64, 1152]> input_93_cast_fp16 = layer_norm(axes = input_93_axes_0, beta = model_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297872128)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307788608)))];
            tensor<fp16, [1, 64, 4304]> linear_46_cast_fp16 = linear(bias = model_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16, x = input_93_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
            tensor<string, []> input_97_mode_0 = const()[name = tensor<string, []>("input_97_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_97_cast_fp16 = gelu(mode = input_97_mode_0, x = linear_46_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307797312)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(317713792)))];
            tensor<fp16, [1, 64, 1152]> linear_47_cast_fp16 = linear(bias = model_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16, x = input_97_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")];
            tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = tensor<string, []>("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(317716160)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(317718528)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = model_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("hidden_states_49_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(317720896)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320375168)))];
            tensor<fp16, [1, 64, 1152]> linear_48_cast_fp16 = linear(bias = model_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320377536)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323031808)))];
            tensor<fp16, [1, 64, 1152]> linear_49_cast_fp16 = linear(bias = model_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323034176)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(325688448)))];
            tensor<fp16, [1, 64, 1152]> linear_50_cast_fp16 = linear(bias = model_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> var_553 = const()[name = tensor<string, []>("op_553"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_554_cast_fp16 = reshape(shape = var_553, x = linear_48_cast_fp16)[name = tensor<string, []>("op_554_cast_fp16")];
            tensor<int32, [4]> var_556 = const()[name = tensor<string, []>("op_556"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_557_cast_fp16 = reshape(shape = var_556, x = linear_49_cast_fp16)[name = tensor<string, []>("op_557_cast_fp16")];
            tensor<int32, [4]> var_559 = const()[name = tensor<string, []>("op_559"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_560_cast_fp16 = reshape(shape = var_559, x = linear_50_cast_fp16)[name = tensor<string, []>("op_560_cast_fp16")];
            tensor<int32, [4]> value_states_35_perm_0 = const()[name = tensor<string, []>("value_states_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_8_y_0_to_fp16 = const()[name = tensor<string, []>("mul_8_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_8_cast_fp16 = mul(x = var_554_cast_fp16, y = mul_8_y_0_to_fp16)[name = tensor<string, []>("mul_8_cast_fp16")];
            tensor<bool, []> matmul_8_transpose_y_0 = const()[name = tensor<string, []>("matmul_8_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_8_transpose_x_0 = const()[name = tensor<string, []>("matmul_8_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_124_perm_0 = const()[name = tensor<string, []>("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_125_perm_0 = const()[name = tensor<string, []>("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_125 = transpose(perm = transpose_125_perm_0, x = var_557_cast_fp16)[name = tensor<string, []>("transpose_235")];
            tensor<fp16, [1, 16, 64, 72]> transpose_124 = transpose(perm = transpose_124_perm_0, x = mul_8_cast_fp16)[name = tensor<string, []>("transpose_236")];
            tensor<fp16, [1, 16, 64, 64]> matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = transpose_124, y = transpose_125)[name = tensor<string, []>("matmul_8_cast_fp16")];
            tensor<int32, []> softmax_8_axis_0 = const()[name = tensor<string, []>("softmax_8_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_8_cast_fp16 = softmax(axis = softmax_8_axis_0, x = matmul_8_cast_fp16)[name = tensor<string, []>("softmax_8_cast_fp16")];
            tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_35_cast_fp16 = transpose(perm = value_states_35_perm_0, x = var_560_cast_fp16)[name = tensor<string, []>("transpose_237")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_33_cast_fp16 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = softmax_8_cast_fp16, y = value_states_35_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")];
            tensor<int32, [4]> var_563_perm_0 = const()[name = tensor<string, []>("op_563_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_565 = const()[name = tensor<string, []>("op_565"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_563_cast_fp16 = transpose(perm = var_563_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_234")];
            tensor<fp16, [1, 64, 1152]> input_101_cast_fp16 = reshape(shape = var_565, x = var_563_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(325690816)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328345088)))];
            tensor<fp16, [1, 64, 1152]> linear_51_cast_fp16 = linear(bias = model_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16, x = input_101_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_103_cast_fp16 = add(x = input_99_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")];
            tensor<int32, [1]> input_105_axes_0 = const()[name = tensor<string, []>("input_105_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328347456)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328349824)))];
            tensor<fp16, [1, 64, 1152]> input_105_cast_fp16 = layer_norm(axes = input_105_axes_0, beta = model_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(328352192)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(338268672)))];
            tensor<fp16, [1, 64, 4304]> linear_52_cast_fp16 = linear(bias = model_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
            tensor<string, []> input_109_mode_0 = const()[name = tensor<string, []>("input_109_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = linear_52_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(338277376)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(348193856)))];
            tensor<fp16, [1, 64, 1152]> linear_53_cast_fp16 = linear(bias = model_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16, x = input_109_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_111_cast_fp16 = add(x = input_103_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")];
            tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(348196224)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(348198592)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = model_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_111_cast_fp16)[name = tensor<string, []>("hidden_states_55_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(348200960)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350855232)))];
            tensor<fp16, [1, 64, 1152]> linear_54_cast_fp16 = linear(bias = model_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350857600)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(353511872)))];
            tensor<fp16, [1, 64, 1152]> linear_55_cast_fp16 = linear(bias = model_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(353514240)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356168512)))];
            tensor<fp16, [1, 64, 1152]> linear_56_cast_fp16 = linear(bias = model_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<int32, [4]> var_608 = const()[name = tensor<string, []>("op_608"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_609_cast_fp16 = reshape(shape = var_608, x = linear_54_cast_fp16)[name = tensor<string, []>("op_609_cast_fp16")];
            tensor<int32, [4]> var_611 = const()[name = tensor<string, []>("op_611"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_612_cast_fp16 = reshape(shape = var_611, x = linear_55_cast_fp16)[name = tensor<string, []>("op_612_cast_fp16")];
            tensor<int32, [4]> var_614 = const()[name = tensor<string, []>("op_614"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_615_cast_fp16 = reshape(shape = var_614, x = linear_56_cast_fp16)[name = tensor<string, []>("op_615_cast_fp16")];
            tensor<int32, [4]> value_states_39_perm_0 = const()[name = tensor<string, []>("value_states_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_9_y_0_to_fp16 = const()[name = tensor<string, []>("mul_9_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_9_cast_fp16 = mul(x = var_609_cast_fp16, y = mul_9_y_0_to_fp16)[name = tensor<string, []>("mul_9_cast_fp16")];
            tensor<bool, []> matmul_9_transpose_y_0 = const()[name = tensor<string, []>("matmul_9_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_9_transpose_x_0 = const()[name = tensor<string, []>("matmul_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_126_perm_0 = const()[name = tensor<string, []>("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_127_perm_0 = const()[name = tensor<string, []>("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_127 = transpose(perm = transpose_127_perm_0, x = var_612_cast_fp16)[name = tensor<string, []>("transpose_231")];
            tensor<fp16, [1, 16, 64, 72]> transpose_126 = transpose(perm = transpose_126_perm_0, x = mul_9_cast_fp16)[name = tensor<string, []>("transpose_232")];
            tensor<fp16, [1, 16, 64, 64]> matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = transpose_126, y = transpose_127)[name = tensor<string, []>("matmul_9_cast_fp16")];
            tensor<int32, []> softmax_9_axis_0 = const()[name = tensor<string, []>("softmax_9_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_9_cast_fp16 = softmax(axis = softmax_9_axis_0, x = matmul_9_cast_fp16)[name = tensor<string, []>("softmax_9_cast_fp16")];
            tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_39_cast_fp16 = transpose(perm = value_states_39_perm_0, x = var_615_cast_fp16)[name = tensor<string, []>("transpose_233")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = softmax_9_cast_fp16, y = value_states_39_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")];
            tensor<int32, [4]> var_618_perm_0 = const()[name = tensor<string, []>("op_618_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_620 = const()[name = tensor<string, []>("op_620"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_618_cast_fp16 = transpose(perm = var_618_perm_0, x = attn_output_37_cast_fp16)[name = tensor<string, []>("transpose_230")];
            tensor<fp16, [1, 64, 1152]> input_113_cast_fp16 = reshape(shape = var_620, x = var_618_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(356170880)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358825152)))];
            tensor<fp16, [1, 64, 1152]> linear_57_cast_fp16 = linear(bias = model_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16, x = input_113_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_115_cast_fp16 = add(x = input_111_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
            tensor<int32, [1]> input_117_axes_0 = const()[name = tensor<string, []>("input_117_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358827520)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358829888)))];
            tensor<fp16, [1, 64, 1152]> input_117_cast_fp16 = layer_norm(axes = input_117_axes_0, beta = model_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358832256)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368748736)))];
            tensor<fp16, [1, 64, 4304]> linear_58_cast_fp16 = linear(bias = model_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16, x = input_117_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
            tensor<string, []> input_121_mode_0 = const()[name = tensor<string, []>("input_121_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_121_cast_fp16 = gelu(mode = input_121_mode_0, x = linear_58_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368757440)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378673920)))];
            tensor<fp16, [1, 64, 1152]> linear_59_cast_fp16 = linear(bias = model_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16, x = input_121_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")];
            tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378676288)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378678656)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = model_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("hidden_states_61_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378681024)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381335296)))];
            tensor<fp16, [1, 64, 1152]> linear_60_cast_fp16 = linear(bias = model_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381337664)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383991936)))];
            tensor<fp16, [1, 64, 1152]> linear_61_cast_fp16 = linear(bias = model_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(383994304)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(386648576)))];
            tensor<fp16, [1, 64, 1152]> linear_62_cast_fp16 = linear(bias = model_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<int32, [4]> var_663 = const()[name = tensor<string, []>("op_663"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_664_cast_fp16 = reshape(shape = var_663, x = linear_60_cast_fp16)[name = tensor<string, []>("op_664_cast_fp16")];
            tensor<int32, [4]> var_666 = const()[name = tensor<string, []>("op_666"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_667_cast_fp16 = reshape(shape = var_666, x = linear_61_cast_fp16)[name = tensor<string, []>("op_667_cast_fp16")];
            tensor<int32, [4]> var_669 = const()[name = tensor<string, []>("op_669"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_670_cast_fp16 = reshape(shape = var_669, x = linear_62_cast_fp16)[name = tensor<string, []>("op_670_cast_fp16")];
            tensor<int32, [4]> value_states_43_perm_0 = const()[name = tensor<string, []>("value_states_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_10_y_0_to_fp16 = const()[name = tensor<string, []>("mul_10_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_10_cast_fp16 = mul(x = var_664_cast_fp16, y = mul_10_y_0_to_fp16)[name = tensor<string, []>("mul_10_cast_fp16")];
            tensor<bool, []> matmul_10_transpose_y_0 = const()[name = tensor<string, []>("matmul_10_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_10_transpose_x_0 = const()[name = tensor<string, []>("matmul_10_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_128_perm_0 = const()[name = tensor<string, []>("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_129_perm_0 = const()[name = tensor<string, []>("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_129 = transpose(perm = transpose_129_perm_0, x = var_667_cast_fp16)[name = tensor<string, []>("transpose_227")];
            tensor<fp16, [1, 16, 64, 72]> transpose_128 = transpose(perm = transpose_128_perm_0, x = mul_10_cast_fp16)[name = tensor<string, []>("transpose_228")];
            tensor<fp16, [1, 16, 64, 64]> matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = transpose_128, y = transpose_129)[name = tensor<string, []>("matmul_10_cast_fp16")];
            tensor<int32, []> softmax_10_axis_0 = const()[name = tensor<string, []>("softmax_10_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_10_cast_fp16 = softmax(axis = softmax_10_axis_0, x = matmul_10_cast_fp16)[name = tensor<string, []>("softmax_10_cast_fp16")];
            tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_43_cast_fp16 = transpose(perm = value_states_43_perm_0, x = var_670_cast_fp16)[name = tensor<string, []>("transpose_229")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_41_cast_fp16 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = softmax_10_cast_fp16, y = value_states_43_cast_fp16)[name = tensor<string, []>("attn_output_41_cast_fp16")];
            tensor<int32, [4]> var_673_perm_0 = const()[name = tensor<string, []>("op_673_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_675 = const()[name = tensor<string, []>("op_675"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_673_cast_fp16 = transpose(perm = var_673_perm_0, x = attn_output_41_cast_fp16)[name = tensor<string, []>("transpose_226")];
            tensor<fp16, [1, 64, 1152]> input_125_cast_fp16 = reshape(shape = var_675, x = var_673_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(386650944)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389305216)))];
            tensor<fp16, [1, 64, 1152]> linear_63_cast_fp16 = linear(bias = model_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16, x = input_125_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_127_cast_fp16 = add(x = input_123_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_127_cast_fp16")];
            tensor<int32, [1]> input_129_axes_0 = const()[name = tensor<string, []>("input_129_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389307584)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389309952)))];
            tensor<fp16, [1, 64, 1152]> input_129_cast_fp16 = layer_norm(axes = input_129_axes_0, beta = model_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_127_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389312320)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(399228800)))];
            tensor<fp16, [1, 64, 4304]> linear_64_cast_fp16 = linear(bias = model_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16, x = input_129_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
            tensor<string, []> input_133_mode_0 = const()[name = tensor<string, []>("input_133_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_133_cast_fp16 = gelu(mode = input_133_mode_0, x = linear_64_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(399237504)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409153984)))];
            tensor<fp16, [1, 64, 1152]> linear_65_cast_fp16 = linear(bias = model_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16, x = input_133_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_135_cast_fp16 = add(x = input_127_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")];
            tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = tensor<string, []>("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409156352)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409158720)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = model_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_135_cast_fp16)[name = tensor<string, []>("hidden_states_67_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409161088)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411815360)))];
            tensor<fp16, [1, 64, 1152]> linear_66_cast_fp16 = linear(bias = model_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411817728)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(414472000)))];
            tensor<fp16, [1, 64, 1152]> linear_67_cast_fp16 = linear(bias = model_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(414474368)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417128640)))];
            tensor<fp16, [1, 64, 1152]> linear_68_cast_fp16 = linear(bias = model_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<int32, [4]> var_718 = const()[name = tensor<string, []>("op_718"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_719_cast_fp16 = reshape(shape = var_718, x = linear_66_cast_fp16)[name = tensor<string, []>("op_719_cast_fp16")];
            tensor<int32, [4]> var_721 = const()[name = tensor<string, []>("op_721"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_722_cast_fp16 = reshape(shape = var_721, x = linear_67_cast_fp16)[name = tensor<string, []>("op_722_cast_fp16")];
            tensor<int32, [4]> var_724 = const()[name = tensor<string, []>("op_724"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_725_cast_fp16 = reshape(shape = var_724, x = linear_68_cast_fp16)[name = tensor<string, []>("op_725_cast_fp16")];
            tensor<int32, [4]> value_states_47_perm_0 = const()[name = tensor<string, []>("value_states_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_11_y_0_to_fp16 = const()[name = tensor<string, []>("mul_11_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_11_cast_fp16 = mul(x = var_719_cast_fp16, y = mul_11_y_0_to_fp16)[name = tensor<string, []>("mul_11_cast_fp16")];
            tensor<bool, []> matmul_11_transpose_y_0 = const()[name = tensor<string, []>("matmul_11_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_11_transpose_x_0 = const()[name = tensor<string, []>("matmul_11_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_130_perm_0 = const()[name = tensor<string, []>("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_131_perm_0 = const()[name = tensor<string, []>("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_131 = transpose(perm = transpose_131_perm_0, x = var_722_cast_fp16)[name = tensor<string, []>("transpose_223")];
            tensor<fp16, [1, 16, 64, 72]> transpose_130 = transpose(perm = transpose_130_perm_0, x = mul_11_cast_fp16)[name = tensor<string, []>("transpose_224")];
            tensor<fp16, [1, 16, 64, 64]> matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = transpose_130, y = transpose_131)[name = tensor<string, []>("matmul_11_cast_fp16")];
            tensor<int32, []> softmax_11_axis_0 = const()[name = tensor<string, []>("softmax_11_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_11_cast_fp16 = softmax(axis = softmax_11_axis_0, x = matmul_11_cast_fp16)[name = tensor<string, []>("softmax_11_cast_fp16")];
            tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_47_cast_fp16 = transpose(perm = value_states_47_perm_0, x = var_725_cast_fp16)[name = tensor<string, []>("transpose_225")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_45_cast_fp16 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = softmax_11_cast_fp16, y = value_states_47_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")];
            tensor<int32, [4]> var_728_perm_0 = const()[name = tensor<string, []>("op_728_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_730 = const()[name = tensor<string, []>("op_730"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_728_cast_fp16 = transpose(perm = var_728_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_222")];
            tensor<fp16, [1, 64, 1152]> input_137_cast_fp16 = reshape(shape = var_730, x = var_728_cast_fp16)[name = tensor<string, []>("input_137_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417131008)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419785280)))];
            tensor<fp16, [1, 64, 1152]> linear_69_cast_fp16 = linear(bias = model_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16, x = input_137_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_139_cast_fp16 = add(x = input_135_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")];
            tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419787648)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419790016)))];
            tensor<fp16, [1, 64, 1152]> input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = model_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(419792384)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429708864)))];
            tensor<fp16, [1, 64, 4304]> linear_70_cast_fp16 = linear(bias = model_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16, x = input_141_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
            tensor<string, []> input_145_mode_0 = const()[name = tensor<string, []>("input_145_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_145_cast_fp16 = gelu(mode = input_145_mode_0, x = linear_70_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429717568)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439634048)))];
            tensor<fp16, [1, 64, 1152]> linear_71_cast_fp16 = linear(bias = model_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16, x = input_145_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")];
            tensor<int32, [1]> hidden_states_73_axes_0 = const()[name = tensor<string, []>("hidden_states_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439636416)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439638784)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_73_cast_fp16 = layer_norm(axes = hidden_states_73_axes_0, beta = model_text_model_encoder_layers_12_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_12_layer_norm1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor<string, []>("hidden_states_73_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439641152)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(442295424)))];
            tensor<fp16, [1, 64, 1152]> linear_72_cast_fp16 = linear(bias = model_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(442297792)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444952064)))];
            tensor<fp16, [1, 64, 1152]> linear_73_cast_fp16 = linear(bias = model_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444954432)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447608704)))];
            tensor<fp16, [1, 64, 1152]> linear_74_cast_fp16 = linear(bias = model_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
            tensor<int32, [4]> var_773 = const()[name = tensor<string, []>("op_773"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_774_cast_fp16 = reshape(shape = var_773, x = linear_72_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, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_777_cast_fp16 = reshape(shape = var_776, x = linear_73_cast_fp16)[name = tensor<string, []>("op_777_cast_fp16")];
            tensor<int32, [4]> var_779 = const()[name = tensor<string, []>("op_779"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_780_cast_fp16 = reshape(shape = var_779, x = linear_74_cast_fp16)[name = tensor<string, []>("op_780_cast_fp16")];
            tensor<int32, [4]> value_states_51_perm_0 = const()[name = tensor<string, []>("value_states_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_12_y_0_to_fp16 = const()[name = tensor<string, []>("mul_12_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_12_cast_fp16 = mul(x = var_774_cast_fp16, y = mul_12_y_0_to_fp16)[name = tensor<string, []>("mul_12_cast_fp16")];
            tensor<bool, []> matmul_12_transpose_y_0 = const()[name = tensor<string, []>("matmul_12_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_12_transpose_x_0 = const()[name = tensor<string, []>("matmul_12_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_132_perm_0 = const()[name = tensor<string, []>("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_133_perm_0 = const()[name = tensor<string, []>("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_133 = transpose(perm = transpose_133_perm_0, x = var_777_cast_fp16)[name = tensor<string, []>("transpose_219")];
            tensor<fp16, [1, 16, 64, 72]> transpose_132 = transpose(perm = transpose_132_perm_0, x = mul_12_cast_fp16)[name = tensor<string, []>("transpose_220")];
            tensor<fp16, [1, 16, 64, 64]> matmul_12_cast_fp16 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = transpose_132, y = transpose_133)[name = tensor<string, []>("matmul_12_cast_fp16")];
            tensor<int32, []> softmax_12_axis_0 = const()[name = tensor<string, []>("softmax_12_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_12_cast_fp16 = softmax(axis = softmax_12_axis_0, x = matmul_12_cast_fp16)[name = tensor<string, []>("softmax_12_cast_fp16")];
            tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_51_cast_fp16 = transpose(perm = value_states_51_perm_0, x = var_780_cast_fp16)[name = tensor<string, []>("transpose_221")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = softmax_12_cast_fp16, y = value_states_51_cast_fp16)[name = tensor<string, []>("attn_output_49_cast_fp16")];
            tensor<int32, [4]> var_783_perm_0 = const()[name = tensor<string, []>("op_783_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_785 = const()[name = tensor<string, []>("op_785"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_783_cast_fp16 = transpose(perm = var_783_perm_0, x = attn_output_49_cast_fp16)[name = tensor<string, []>("transpose_218")];
            tensor<fp16, [1, 64, 1152]> input_149_cast_fp16 = reshape(shape = var_785, x = var_783_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447611072)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450265344)))];
            tensor<fp16, [1, 64, 1152]> linear_75_cast_fp16 = linear(bias = model_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16, x = input_149_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_151_cast_fp16 = add(x = input_147_cast_fp16, y = linear_75_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")];
            tensor<int32, [1]> input_153_axes_0 = const()[name = tensor<string, []>("input_153_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450267712)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450270080)))];
            tensor<fp16, [1, 64, 1152]> input_153_cast_fp16 = layer_norm(axes = input_153_axes_0, beta = model_text_model_encoder_layers_12_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_12_layer_norm2_weight_to_fp16, x = input_151_cast_fp16)[name = tensor<string, []>("input_153_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450272448)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460188928)))];
            tensor<fp16, [1, 64, 4304]> linear_76_cast_fp16 = linear(bias = model_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16, x = input_153_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")];
            tensor<string, []> input_157_mode_0 = const()[name = tensor<string, []>("input_157_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_157_cast_fp16 = gelu(mode = input_157_mode_0, x = linear_76_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460197632)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470114112)))];
            tensor<fp16, [1, 64, 1152]> linear_77_cast_fp16 = linear(bias = model_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16, x = input_157_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_159_cast_fp16 = add(x = input_151_cast_fp16, y = linear_77_cast_fp16)[name = tensor<string, []>("input_159_cast_fp16")];
            tensor<int32, [1]> hidden_states_79_axes_0 = const()[name = tensor<string, []>("hidden_states_79_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470116480)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470118848)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_79_cast_fp16 = layer_norm(axes = hidden_states_79_axes_0, beta = model_text_model_encoder_layers_13_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_13_layer_norm1_weight_to_fp16, x = input_159_cast_fp16)[name = tensor<string, []>("hidden_states_79_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470121216)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(472775488)))];
            tensor<fp16, [1, 64, 1152]> linear_78_cast_fp16 = linear(bias = model_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_78_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(472777856)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(475432128)))];
            tensor<fp16, [1, 64, 1152]> linear_79_cast_fp16 = linear(bias = model_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(475434496)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478088768)))];
            tensor<fp16, [1, 64, 1152]> linear_80_cast_fp16 = linear(bias = model_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")];
            tensor<int32, [4]> var_828 = const()[name = tensor<string, []>("op_828"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_829_cast_fp16 = reshape(shape = var_828, x = linear_78_cast_fp16)[name = tensor<string, []>("op_829_cast_fp16")];
            tensor<int32, [4]> var_831 = const()[name = tensor<string, []>("op_831"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_832_cast_fp16 = reshape(shape = var_831, x = linear_79_cast_fp16)[name = tensor<string, []>("op_832_cast_fp16")];
            tensor<int32, [4]> var_834 = const()[name = tensor<string, []>("op_834"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_835_cast_fp16 = reshape(shape = var_834, x = linear_80_cast_fp16)[name = tensor<string, []>("op_835_cast_fp16")];
            tensor<int32, [4]> value_states_55_perm_0 = const()[name = tensor<string, []>("value_states_55_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_13_y_0_to_fp16 = const()[name = tensor<string, []>("mul_13_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_13_cast_fp16 = mul(x = var_829_cast_fp16, y = mul_13_y_0_to_fp16)[name = tensor<string, []>("mul_13_cast_fp16")];
            tensor<bool, []> matmul_13_transpose_y_0 = const()[name = tensor<string, []>("matmul_13_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_13_transpose_x_0 = const()[name = tensor<string, []>("matmul_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_134_perm_0 = const()[name = tensor<string, []>("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_135_perm_0 = const()[name = tensor<string, []>("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_135 = transpose(perm = transpose_135_perm_0, x = var_832_cast_fp16)[name = tensor<string, []>("transpose_215")];
            tensor<fp16, [1, 16, 64, 72]> transpose_134 = transpose(perm = transpose_134_perm_0, x = mul_13_cast_fp16)[name = tensor<string, []>("transpose_216")];
            tensor<fp16, [1, 16, 64, 64]> matmul_13_cast_fp16 = matmul(transpose_x = matmul_13_transpose_x_0, transpose_y = matmul_13_transpose_y_0, x = transpose_134, y = transpose_135)[name = tensor<string, []>("matmul_13_cast_fp16")];
            tensor<int32, []> softmax_13_axis_0 = const()[name = tensor<string, []>("softmax_13_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_13_cast_fp16 = softmax(axis = softmax_13_axis_0, x = matmul_13_cast_fp16)[name = tensor<string, []>("softmax_13_cast_fp16")];
            tensor<bool, []> attn_output_53_transpose_x_0 = const()[name = tensor<string, []>("attn_output_53_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_53_transpose_y_0 = const()[name = tensor<string, []>("attn_output_53_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_55_cast_fp16 = transpose(perm = value_states_55_perm_0, x = var_835_cast_fp16)[name = tensor<string, []>("transpose_217")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_53_cast_fp16 = matmul(transpose_x = attn_output_53_transpose_x_0, transpose_y = attn_output_53_transpose_y_0, x = softmax_13_cast_fp16, y = value_states_55_cast_fp16)[name = tensor<string, []>("attn_output_53_cast_fp16")];
            tensor<int32, [4]> var_838_perm_0 = const()[name = tensor<string, []>("op_838_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_840 = const()[name = tensor<string, []>("op_840"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_838_cast_fp16 = transpose(perm = var_838_perm_0, x = attn_output_53_cast_fp16)[name = tensor<string, []>("transpose_214")];
            tensor<fp16, [1, 64, 1152]> input_161_cast_fp16 = reshape(shape = var_840, x = var_838_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478091136)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480745408)))];
            tensor<fp16, [1, 64, 1152]> linear_81_cast_fp16 = linear(bias = model_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16, x = input_161_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_163_cast_fp16 = add(x = input_159_cast_fp16, y = linear_81_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")];
            tensor<int32, [1]> input_165_axes_0 = const()[name = tensor<string, []>("input_165_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480747776)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480750144)))];
            tensor<fp16, [1, 64, 1152]> input_165_cast_fp16 = layer_norm(axes = input_165_axes_0, beta = model_text_model_encoder_layers_13_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_13_layer_norm2_weight_to_fp16, x = input_163_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480752512)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490668992)))];
            tensor<fp16, [1, 64, 4304]> linear_82_cast_fp16 = linear(bias = model_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16, x = input_165_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")];
            tensor<string, []> input_169_mode_0 = const()[name = tensor<string, []>("input_169_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_169_cast_fp16 = gelu(mode = input_169_mode_0, x = linear_82_cast_fp16)[name = tensor<string, []>("input_169_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490677696)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500594176)))];
            tensor<fp16, [1, 64, 1152]> linear_83_cast_fp16 = linear(bias = model_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16, x = input_169_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_83_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")];
            tensor<int32, [1]> hidden_states_85_axes_0 = const()[name = tensor<string, []>("hidden_states_85_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500596544)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500598912)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_85_cast_fp16 = layer_norm(axes = hidden_states_85_axes_0, beta = model_text_model_encoder_layers_14_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_14_layer_norm1_weight_to_fp16, x = input_171_cast_fp16)[name = tensor<string, []>("hidden_states_85_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500601280)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(503255552)))];
            tensor<fp16, [1, 64, 1152]> linear_84_cast_fp16 = linear(bias = model_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(503257920)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505912192)))];
            tensor<fp16, [1, 64, 1152]> linear_85_cast_fp16 = linear(bias = model_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505914560)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508568832)))];
            tensor<fp16, [1, 64, 1152]> linear_86_cast_fp16 = linear(bias = model_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")];
            tensor<int32, [4]> var_883 = const()[name = tensor<string, []>("op_883"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_884_cast_fp16 = reshape(shape = var_883, x = linear_84_cast_fp16)[name = tensor<string, []>("op_884_cast_fp16")];
            tensor<int32, [4]> var_886 = const()[name = tensor<string, []>("op_886"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_887_cast_fp16 = reshape(shape = var_886, x = linear_85_cast_fp16)[name = tensor<string, []>("op_887_cast_fp16")];
            tensor<int32, [4]> var_889 = const()[name = tensor<string, []>("op_889"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_890_cast_fp16 = reshape(shape = var_889, x = linear_86_cast_fp16)[name = tensor<string, []>("op_890_cast_fp16")];
            tensor<int32, [4]> value_states_59_perm_0 = const()[name = tensor<string, []>("value_states_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_14_y_0_to_fp16 = const()[name = tensor<string, []>("mul_14_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_14_cast_fp16 = mul(x = var_884_cast_fp16, y = mul_14_y_0_to_fp16)[name = tensor<string, []>("mul_14_cast_fp16")];
            tensor<bool, []> matmul_14_transpose_y_0 = const()[name = tensor<string, []>("matmul_14_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_14_transpose_x_0 = const()[name = tensor<string, []>("matmul_14_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_136_perm_0 = const()[name = tensor<string, []>("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_137_perm_0 = const()[name = tensor<string, []>("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_137 = transpose(perm = transpose_137_perm_0, x = var_887_cast_fp16)[name = tensor<string, []>("transpose_211")];
            tensor<fp16, [1, 16, 64, 72]> transpose_136 = transpose(perm = transpose_136_perm_0, x = mul_14_cast_fp16)[name = tensor<string, []>("transpose_212")];
            tensor<fp16, [1, 16, 64, 64]> matmul_14_cast_fp16 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = transpose_136, y = transpose_137)[name = tensor<string, []>("matmul_14_cast_fp16")];
            tensor<int32, []> softmax_14_axis_0 = const()[name = tensor<string, []>("softmax_14_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_14_cast_fp16 = softmax(axis = softmax_14_axis_0, x = matmul_14_cast_fp16)[name = tensor<string, []>("softmax_14_cast_fp16")];
            tensor<bool, []> attn_output_57_transpose_x_0 = const()[name = tensor<string, []>("attn_output_57_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_57_transpose_y_0 = const()[name = tensor<string, []>("attn_output_57_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_59_cast_fp16 = transpose(perm = value_states_59_perm_0, x = var_890_cast_fp16)[name = tensor<string, []>("transpose_213")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_57_cast_fp16 = matmul(transpose_x = attn_output_57_transpose_x_0, transpose_y = attn_output_57_transpose_y_0, x = softmax_14_cast_fp16, y = value_states_59_cast_fp16)[name = tensor<string, []>("attn_output_57_cast_fp16")];
            tensor<int32, [4]> var_893_perm_0 = const()[name = tensor<string, []>("op_893_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_895 = const()[name = tensor<string, []>("op_895"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_893_cast_fp16 = transpose(perm = var_893_perm_0, x = attn_output_57_cast_fp16)[name = tensor<string, []>("transpose_210")];
            tensor<fp16, [1, 64, 1152]> input_173_cast_fp16 = reshape(shape = var_895, x = var_893_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508571200)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(511225472)))];
            tensor<fp16, [1, 64, 1152]> linear_87_cast_fp16 = linear(bias = model_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16, x = input_173_cast_fp16)[name = tensor<string, []>("linear_87_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_175_cast_fp16 = add(x = input_171_cast_fp16, y = linear_87_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
            tensor<int32, [1]> input_177_axes_0 = const()[name = tensor<string, []>("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(511227840)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(511230208)))];
            tensor<fp16, [1, 64, 1152]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = model_text_model_encoder_layers_14_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_14_layer_norm2_weight_to_fp16, x = input_175_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(511232576)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(521149056)))];
            tensor<fp16, [1, 64, 4304]> linear_88_cast_fp16 = linear(bias = model_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16, x = input_177_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")];
            tensor<string, []> input_181_mode_0 = const()[name = tensor<string, []>("input_181_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_181_cast_fp16 = gelu(mode = input_181_mode_0, x = linear_88_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(521157760)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(531074240)))];
            tensor<fp16, [1, 64, 1152]> linear_89_cast_fp16 = linear(bias = model_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16, x = input_181_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_183_cast_fp16 = add(x = input_175_cast_fp16, y = linear_89_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")];
            tensor<int32, [1]> hidden_states_91_axes_0 = const()[name = tensor<string, []>("hidden_states_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(531076608)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(531078976)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_91_cast_fp16 = layer_norm(axes = hidden_states_91_axes_0, beta = model_text_model_encoder_layers_15_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_15_layer_norm1_weight_to_fp16, x = input_183_cast_fp16)[name = tensor<string, []>("hidden_states_91_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(531081344)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533735616)))];
            tensor<fp16, [1, 64, 1152]> linear_90_cast_fp16 = linear(bias = model_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533737984)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(536392256)))];
            tensor<fp16, [1, 64, 1152]> linear_91_cast_fp16 = linear(bias = model_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(536394624)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539048896)))];
            tensor<fp16, [1, 64, 1152]> linear_92_cast_fp16 = linear(bias = model_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")];
            tensor<int32, [4]> var_938 = const()[name = tensor<string, []>("op_938"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_939_cast_fp16 = reshape(shape = var_938, x = linear_90_cast_fp16)[name = tensor<string, []>("op_939_cast_fp16")];
            tensor<int32, [4]> var_941 = const()[name = tensor<string, []>("op_941"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_942_cast_fp16 = reshape(shape = var_941, x = linear_91_cast_fp16)[name = tensor<string, []>("op_942_cast_fp16")];
            tensor<int32, [4]> var_944 = const()[name = tensor<string, []>("op_944"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_945_cast_fp16 = reshape(shape = var_944, x = linear_92_cast_fp16)[name = tensor<string, []>("op_945_cast_fp16")];
            tensor<int32, [4]> value_states_63_perm_0 = const()[name = tensor<string, []>("value_states_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_15_y_0_to_fp16 = const()[name = tensor<string, []>("mul_15_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_15_cast_fp16 = mul(x = var_939_cast_fp16, y = mul_15_y_0_to_fp16)[name = tensor<string, []>("mul_15_cast_fp16")];
            tensor<bool, []> matmul_15_transpose_y_0 = const()[name = tensor<string, []>("matmul_15_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_15_transpose_x_0 = const()[name = tensor<string, []>("matmul_15_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_138_perm_0 = const()[name = tensor<string, []>("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_139_perm_0 = const()[name = tensor<string, []>("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_139 = transpose(perm = transpose_139_perm_0, x = var_942_cast_fp16)[name = tensor<string, []>("transpose_207")];
            tensor<fp16, [1, 16, 64, 72]> transpose_138 = transpose(perm = transpose_138_perm_0, x = mul_15_cast_fp16)[name = tensor<string, []>("transpose_208")];
            tensor<fp16, [1, 16, 64, 64]> matmul_15_cast_fp16 = matmul(transpose_x = matmul_15_transpose_x_0, transpose_y = matmul_15_transpose_y_0, x = transpose_138, y = transpose_139)[name = tensor<string, []>("matmul_15_cast_fp16")];
            tensor<int32, []> softmax_15_axis_0 = const()[name = tensor<string, []>("softmax_15_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_15_cast_fp16 = softmax(axis = softmax_15_axis_0, x = matmul_15_cast_fp16)[name = tensor<string, []>("softmax_15_cast_fp16")];
            tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_63_cast_fp16 = transpose(perm = value_states_63_perm_0, x = var_945_cast_fp16)[name = tensor<string, []>("transpose_209")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = softmax_15_cast_fp16, y = value_states_63_cast_fp16)[name = tensor<string, []>("attn_output_61_cast_fp16")];
            tensor<int32, [4]> var_948_perm_0 = const()[name = tensor<string, []>("op_948_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_950 = const()[name = tensor<string, []>("op_950"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_948_cast_fp16 = transpose(perm = var_948_perm_0, x = attn_output_61_cast_fp16)[name = tensor<string, []>("transpose_206")];
            tensor<fp16, [1, 64, 1152]> input_185_cast_fp16 = reshape(shape = var_950, x = var_948_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539051264)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541705536)))];
            tensor<fp16, [1, 64, 1152]> linear_93_cast_fp16 = linear(bias = model_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16, x = input_185_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_187_cast_fp16 = add(x = input_183_cast_fp16, y = linear_93_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")];
            tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541707904)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541710272)))];
            tensor<fp16, [1, 64, 1152]> input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = model_text_model_encoder_layers_15_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_15_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(541712640)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(551629120)))];
            tensor<fp16, [1, 64, 4304]> linear_94_cast_fp16 = linear(bias = model_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16, x = input_189_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")];
            tensor<string, []> input_193_mode_0 = const()[name = tensor<string, []>("input_193_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_193_cast_fp16 = gelu(mode = input_193_mode_0, x = linear_94_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(551637824)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(561554304)))];
            tensor<fp16, [1, 64, 1152]> linear_95_cast_fp16 = linear(bias = model_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16, x = input_193_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_95_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")];
            tensor<int32, [1]> hidden_states_97_axes_0 = const()[name = tensor<string, []>("hidden_states_97_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(561556672)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(561559040)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_97_cast_fp16 = layer_norm(axes = hidden_states_97_axes_0, beta = model_text_model_encoder_layers_16_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_16_layer_norm1_weight_to_fp16, x = input_195_cast_fp16)[name = tensor<string, []>("hidden_states_97_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(561561408)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564215680)))];
            tensor<fp16, [1, 64, 1152]> linear_96_cast_fp16 = linear(bias = model_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_96_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564218048)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566872320)))];
            tensor<fp16, [1, 64, 1152]> linear_97_cast_fp16 = linear(bias = model_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566874688)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(569528960)))];
            tensor<fp16, [1, 64, 1152]> linear_98_cast_fp16 = linear(bias = model_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")];
            tensor<int32, [4]> var_993 = const()[name = tensor<string, []>("op_993"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_994_cast_fp16 = reshape(shape = var_993, x = linear_96_cast_fp16)[name = tensor<string, []>("op_994_cast_fp16")];
            tensor<int32, [4]> var_996 = const()[name = tensor<string, []>("op_996"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_997_cast_fp16 = reshape(shape = var_996, x = linear_97_cast_fp16)[name = tensor<string, []>("op_997_cast_fp16")];
            tensor<int32, [4]> var_999 = const()[name = tensor<string, []>("op_999"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1000_cast_fp16 = reshape(shape = var_999, x = linear_98_cast_fp16)[name = tensor<string, []>("op_1000_cast_fp16")];
            tensor<int32, [4]> value_states_67_perm_0 = const()[name = tensor<string, []>("value_states_67_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_16_y_0_to_fp16 = const()[name = tensor<string, []>("mul_16_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_16_cast_fp16 = mul(x = var_994_cast_fp16, y = mul_16_y_0_to_fp16)[name = tensor<string, []>("mul_16_cast_fp16")];
            tensor<bool, []> matmul_16_transpose_y_0 = const()[name = tensor<string, []>("matmul_16_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_16_transpose_x_0 = const()[name = tensor<string, []>("matmul_16_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_140_perm_0 = const()[name = tensor<string, []>("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_141_perm_0 = const()[name = tensor<string, []>("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_141 = transpose(perm = transpose_141_perm_0, x = var_997_cast_fp16)[name = tensor<string, []>("transpose_203")];
            tensor<fp16, [1, 16, 64, 72]> transpose_140 = transpose(perm = transpose_140_perm_0, x = mul_16_cast_fp16)[name = tensor<string, []>("transpose_204")];
            tensor<fp16, [1, 16, 64, 64]> matmul_16_cast_fp16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = transpose_140, y = transpose_141)[name = tensor<string, []>("matmul_16_cast_fp16")];
            tensor<int32, []> softmax_16_axis_0 = const()[name = tensor<string, []>("softmax_16_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_16_cast_fp16 = softmax(axis = softmax_16_axis_0, x = matmul_16_cast_fp16)[name = tensor<string, []>("softmax_16_cast_fp16")];
            tensor<bool, []> attn_output_65_transpose_x_0 = const()[name = tensor<string, []>("attn_output_65_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_65_transpose_y_0 = const()[name = tensor<string, []>("attn_output_65_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_67_cast_fp16 = transpose(perm = value_states_67_perm_0, x = var_1000_cast_fp16)[name = tensor<string, []>("transpose_205")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_65_cast_fp16 = matmul(transpose_x = attn_output_65_transpose_x_0, transpose_y = attn_output_65_transpose_y_0, x = softmax_16_cast_fp16, y = value_states_67_cast_fp16)[name = tensor<string, []>("attn_output_65_cast_fp16")];
            tensor<int32, [4]> var_1003_perm_0 = const()[name = tensor<string, []>("op_1003_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1005 = const()[name = tensor<string, []>("op_1005"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1003_cast_fp16 = transpose(perm = var_1003_perm_0, x = attn_output_65_cast_fp16)[name = tensor<string, []>("transpose_202")];
            tensor<fp16, [1, 64, 1152]> input_197_cast_fp16 = reshape(shape = var_1005, x = var_1003_cast_fp16)[name = tensor<string, []>("input_197_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(569531328)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572185600)))];
            tensor<fp16, [1, 64, 1152]> linear_99_cast_fp16 = linear(bias = model_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16, x = input_197_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_199_cast_fp16 = add(x = input_195_cast_fp16, y = linear_99_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")];
            tensor<int32, [1]> input_201_axes_0 = const()[name = tensor<string, []>("input_201_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572187968)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572190336)))];
            tensor<fp16, [1, 64, 1152]> input_201_cast_fp16 = layer_norm(axes = input_201_axes_0, beta = model_text_model_encoder_layers_16_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_16_layer_norm2_weight_to_fp16, x = input_199_cast_fp16)[name = tensor<string, []>("input_201_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572192704)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(582109184)))];
            tensor<fp16, [1, 64, 4304]> linear_100_cast_fp16 = linear(bias = model_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16, x = input_201_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")];
            tensor<string, []> input_205_mode_0 = const()[name = tensor<string, []>("input_205_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_205_cast_fp16 = gelu(mode = input_205_mode_0, x = linear_100_cast_fp16)[name = tensor<string, []>("input_205_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(582117888)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592034368)))];
            tensor<fp16, [1, 64, 1152]> linear_101_cast_fp16 = linear(bias = model_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16, x = input_205_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_207_cast_fp16 = add(x = input_199_cast_fp16, y = linear_101_cast_fp16)[name = tensor<string, []>("input_207_cast_fp16")];
            tensor<int32, [1]> hidden_states_103_axes_0 = const()[name = tensor<string, []>("hidden_states_103_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592036736)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592039104)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_103_cast_fp16 = layer_norm(axes = hidden_states_103_axes_0, beta = model_text_model_encoder_layers_17_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_17_layer_norm1_weight_to_fp16, x = input_207_cast_fp16)[name = tensor<string, []>("hidden_states_103_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(592041472)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(594695744)))];
            tensor<fp16, [1, 64, 1152]> linear_102_cast_fp16 = linear(bias = model_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(594698112)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597352384)))];
            tensor<fp16, [1, 64, 1152]> linear_103_cast_fp16 = linear(bias = model_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(597354752)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(600009024)))];
            tensor<fp16, [1, 64, 1152]> linear_104_cast_fp16 = linear(bias = model_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")];
            tensor<int32, [4]> var_1048 = const()[name = tensor<string, []>("op_1048"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1049_cast_fp16 = reshape(shape = var_1048, x = linear_102_cast_fp16)[name = tensor<string, []>("op_1049_cast_fp16")];
            tensor<int32, [4]> var_1051 = const()[name = tensor<string, []>("op_1051"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1052_cast_fp16 = reshape(shape = var_1051, x = linear_103_cast_fp16)[name = tensor<string, []>("op_1052_cast_fp16")];
            tensor<int32, [4]> var_1054 = const()[name = tensor<string, []>("op_1054"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1055_cast_fp16 = reshape(shape = var_1054, x = linear_104_cast_fp16)[name = tensor<string, []>("op_1055_cast_fp16")];
            tensor<int32, [4]> value_states_71_perm_0 = const()[name = tensor<string, []>("value_states_71_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_17_y_0_to_fp16 = const()[name = tensor<string, []>("mul_17_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_17_cast_fp16 = mul(x = var_1049_cast_fp16, y = mul_17_y_0_to_fp16)[name = tensor<string, []>("mul_17_cast_fp16")];
            tensor<bool, []> matmul_17_transpose_y_0 = const()[name = tensor<string, []>("matmul_17_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_17_transpose_x_0 = const()[name = tensor<string, []>("matmul_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_142_perm_0 = const()[name = tensor<string, []>("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_143_perm_0 = const()[name = tensor<string, []>("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_143 = transpose(perm = transpose_143_perm_0, x = var_1052_cast_fp16)[name = tensor<string, []>("transpose_199")];
            tensor<fp16, [1, 16, 64, 72]> transpose_142 = transpose(perm = transpose_142_perm_0, x = mul_17_cast_fp16)[name = tensor<string, []>("transpose_200")];
            tensor<fp16, [1, 16, 64, 64]> matmul_17_cast_fp16 = matmul(transpose_x = matmul_17_transpose_x_0, transpose_y = matmul_17_transpose_y_0, x = transpose_142, y = transpose_143)[name = tensor<string, []>("matmul_17_cast_fp16")];
            tensor<int32, []> softmax_17_axis_0 = const()[name = tensor<string, []>("softmax_17_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_17_cast_fp16 = softmax(axis = softmax_17_axis_0, x = matmul_17_cast_fp16)[name = tensor<string, []>("softmax_17_cast_fp16")];
            tensor<bool, []> attn_output_69_transpose_x_0 = const()[name = tensor<string, []>("attn_output_69_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_69_transpose_y_0 = const()[name = tensor<string, []>("attn_output_69_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_71_cast_fp16 = transpose(perm = value_states_71_perm_0, x = var_1055_cast_fp16)[name = tensor<string, []>("transpose_201")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_69_cast_fp16 = matmul(transpose_x = attn_output_69_transpose_x_0, transpose_y = attn_output_69_transpose_y_0, x = softmax_17_cast_fp16, y = value_states_71_cast_fp16)[name = tensor<string, []>("attn_output_69_cast_fp16")];
            tensor<int32, [4]> var_1058_perm_0 = const()[name = tensor<string, []>("op_1058_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1060 = const()[name = tensor<string, []>("op_1060"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1058_cast_fp16 = transpose(perm = var_1058_perm_0, x = attn_output_69_cast_fp16)[name = tensor<string, []>("transpose_198")];
            tensor<fp16, [1, 64, 1152]> input_209_cast_fp16 = reshape(shape = var_1060, x = var_1058_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(600011392)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602665664)))];
            tensor<fp16, [1, 64, 1152]> linear_105_cast_fp16 = linear(bias = model_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16, x = input_209_cast_fp16)[name = tensor<string, []>("linear_105_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_211_cast_fp16 = add(x = input_207_cast_fp16, y = linear_105_cast_fp16)[name = tensor<string, []>("input_211_cast_fp16")];
            tensor<int32, [1]> input_213_axes_0 = const()[name = tensor<string, []>("input_213_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602668032)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602670400)))];
            tensor<fp16, [1, 64, 1152]> input_213_cast_fp16 = layer_norm(axes = input_213_axes_0, beta = model_text_model_encoder_layers_17_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_17_layer_norm2_weight_to_fp16, x = input_211_cast_fp16)[name = tensor<string, []>("input_213_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602672768)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(612589248)))];
            tensor<fp16, [1, 64, 4304]> linear_106_cast_fp16 = linear(bias = model_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16, x = input_213_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")];
            tensor<string, []> input_217_mode_0 = const()[name = tensor<string, []>("input_217_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_217_cast_fp16 = gelu(mode = input_217_mode_0, x = linear_106_cast_fp16)[name = tensor<string, []>("input_217_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(612597952)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622514432)))];
            tensor<fp16, [1, 64, 1152]> linear_107_cast_fp16 = linear(bias = model_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16, x = input_217_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_219_cast_fp16 = add(x = input_211_cast_fp16, y = linear_107_cast_fp16)[name = tensor<string, []>("input_219_cast_fp16")];
            tensor<int32, [1]> hidden_states_109_axes_0 = const()[name = tensor<string, []>("hidden_states_109_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622516800)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622519168)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_109_cast_fp16 = layer_norm(axes = hidden_states_109_axes_0, beta = model_text_model_encoder_layers_18_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_18_layer_norm1_weight_to_fp16, x = input_219_cast_fp16)[name = tensor<string, []>("hidden_states_109_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(622521536)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625175808)))];
            tensor<fp16, [1, 64, 1152]> linear_108_cast_fp16 = linear(bias = model_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625178176)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(627832448)))];
            tensor<fp16, [1, 64, 1152]> linear_109_cast_fp16 = linear(bias = model_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(627834816)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(630489088)))];
            tensor<fp16, [1, 64, 1152]> linear_110_cast_fp16 = linear(bias = model_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")];
            tensor<int32, [4]> var_1103 = const()[name = tensor<string, []>("op_1103"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1104_cast_fp16 = reshape(shape = var_1103, x = linear_108_cast_fp16)[name = tensor<string, []>("op_1104_cast_fp16")];
            tensor<int32, [4]> var_1106 = const()[name = tensor<string, []>("op_1106"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1107_cast_fp16 = reshape(shape = var_1106, x = linear_109_cast_fp16)[name = tensor<string, []>("op_1107_cast_fp16")];
            tensor<int32, [4]> var_1109 = const()[name = tensor<string, []>("op_1109"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1110_cast_fp16 = reshape(shape = var_1109, x = linear_110_cast_fp16)[name = tensor<string, []>("op_1110_cast_fp16")];
            tensor<int32, [4]> value_states_75_perm_0 = const()[name = tensor<string, []>("value_states_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_18_y_0_to_fp16 = const()[name = tensor<string, []>("mul_18_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_18_cast_fp16 = mul(x = var_1104_cast_fp16, y = mul_18_y_0_to_fp16)[name = tensor<string, []>("mul_18_cast_fp16")];
            tensor<bool, []> matmul_18_transpose_y_0 = const()[name = tensor<string, []>("matmul_18_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_18_transpose_x_0 = const()[name = tensor<string, []>("matmul_18_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_144_perm_0 = const()[name = tensor<string, []>("transpose_144_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_145_perm_0 = const()[name = tensor<string, []>("transpose_145_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_145 = transpose(perm = transpose_145_perm_0, x = var_1107_cast_fp16)[name = tensor<string, []>("transpose_195")];
            tensor<fp16, [1, 16, 64, 72]> transpose_144 = transpose(perm = transpose_144_perm_0, x = mul_18_cast_fp16)[name = tensor<string, []>("transpose_196")];
            tensor<fp16, [1, 16, 64, 64]> matmul_18_cast_fp16 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = transpose_144, y = transpose_145)[name = tensor<string, []>("matmul_18_cast_fp16")];
            tensor<int32, []> softmax_18_axis_0 = const()[name = tensor<string, []>("softmax_18_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_18_cast_fp16 = softmax(axis = softmax_18_axis_0, x = matmul_18_cast_fp16)[name = tensor<string, []>("softmax_18_cast_fp16")];
            tensor<bool, []> attn_output_73_transpose_x_0 = const()[name = tensor<string, []>("attn_output_73_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_73_transpose_y_0 = const()[name = tensor<string, []>("attn_output_73_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_75_cast_fp16 = transpose(perm = value_states_75_perm_0, x = var_1110_cast_fp16)[name = tensor<string, []>("transpose_197")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_73_cast_fp16 = matmul(transpose_x = attn_output_73_transpose_x_0, transpose_y = attn_output_73_transpose_y_0, x = softmax_18_cast_fp16, y = value_states_75_cast_fp16)[name = tensor<string, []>("attn_output_73_cast_fp16")];
            tensor<int32, [4]> var_1113_perm_0 = const()[name = tensor<string, []>("op_1113_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1115 = const()[name = tensor<string, []>("op_1115"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1113_cast_fp16 = transpose(perm = var_1113_perm_0, x = attn_output_73_cast_fp16)[name = tensor<string, []>("transpose_194")];
            tensor<fp16, [1, 64, 1152]> input_221_cast_fp16 = reshape(shape = var_1115, x = var_1113_cast_fp16)[name = tensor<string, []>("input_221_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(630491456)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633145728)))];
            tensor<fp16, [1, 64, 1152]> linear_111_cast_fp16 = linear(bias = model_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16, x = input_221_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_223_cast_fp16 = add(x = input_219_cast_fp16, y = linear_111_cast_fp16)[name = tensor<string, []>("input_223_cast_fp16")];
            tensor<int32, [1]> input_225_axes_0 = const()[name = tensor<string, []>("input_225_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633148096)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633150464)))];
            tensor<fp16, [1, 64, 1152]> input_225_cast_fp16 = layer_norm(axes = input_225_axes_0, beta = model_text_model_encoder_layers_18_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_18_layer_norm2_weight_to_fp16, x = input_223_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(633152832)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643069312)))];
            tensor<fp16, [1, 64, 4304]> linear_112_cast_fp16 = linear(bias = model_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16, x = input_225_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")];
            tensor<string, []> input_229_mode_0 = const()[name = tensor<string, []>("input_229_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_229_cast_fp16 = gelu(mode = input_229_mode_0, x = linear_112_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643078016)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652994496)))];
            tensor<fp16, [1, 64, 1152]> linear_113_cast_fp16 = linear(bias = model_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16, x = input_229_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_231_cast_fp16 = add(x = input_223_cast_fp16, y = linear_113_cast_fp16)[name = tensor<string, []>("input_231_cast_fp16")];
            tensor<int32, [1]> hidden_states_115_axes_0 = const()[name = tensor<string, []>("hidden_states_115_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652996864)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(652999232)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_115_cast_fp16 = layer_norm(axes = hidden_states_115_axes_0, beta = model_text_model_encoder_layers_19_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_19_layer_norm1_weight_to_fp16, x = input_231_cast_fp16)[name = tensor<string, []>("hidden_states_115_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(653001600)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(655655872)))];
            tensor<fp16, [1, 64, 1152]> linear_114_cast_fp16 = linear(bias = model_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_114_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(655658240)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(658312512)))];
            tensor<fp16, [1, 64, 1152]> linear_115_cast_fp16 = linear(bias = model_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(658314880)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(660969152)))];
            tensor<fp16, [1, 64, 1152]> linear_116_cast_fp16 = linear(bias = model_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")];
            tensor<int32, [4]> var_1158 = const()[name = tensor<string, []>("op_1158"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1159_cast_fp16 = reshape(shape = var_1158, x = linear_114_cast_fp16)[name = tensor<string, []>("op_1159_cast_fp16")];
            tensor<int32, [4]> var_1161 = const()[name = tensor<string, []>("op_1161"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1162_cast_fp16 = reshape(shape = var_1161, x = linear_115_cast_fp16)[name = tensor<string, []>("op_1162_cast_fp16")];
            tensor<int32, [4]> var_1164 = const()[name = tensor<string, []>("op_1164"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1165_cast_fp16 = reshape(shape = var_1164, x = linear_116_cast_fp16)[name = tensor<string, []>("op_1165_cast_fp16")];
            tensor<int32, [4]> value_states_79_perm_0 = const()[name = tensor<string, []>("value_states_79_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_19_y_0_to_fp16 = const()[name = tensor<string, []>("mul_19_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_19_cast_fp16 = mul(x = var_1159_cast_fp16, y = mul_19_y_0_to_fp16)[name = tensor<string, []>("mul_19_cast_fp16")];
            tensor<bool, []> matmul_19_transpose_y_0 = const()[name = tensor<string, []>("matmul_19_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_19_transpose_x_0 = const()[name = tensor<string, []>("matmul_19_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_146_perm_0 = const()[name = tensor<string, []>("transpose_146_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_147_perm_0 = const()[name = tensor<string, []>("transpose_147_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_147 = transpose(perm = transpose_147_perm_0, x = var_1162_cast_fp16)[name = tensor<string, []>("transpose_191")];
            tensor<fp16, [1, 16, 64, 72]> transpose_146 = transpose(perm = transpose_146_perm_0, x = mul_19_cast_fp16)[name = tensor<string, []>("transpose_192")];
            tensor<fp16, [1, 16, 64, 64]> matmul_19_cast_fp16 = matmul(transpose_x = matmul_19_transpose_x_0, transpose_y = matmul_19_transpose_y_0, x = transpose_146, y = transpose_147)[name = tensor<string, []>("matmul_19_cast_fp16")];
            tensor<int32, []> softmax_19_axis_0 = const()[name = tensor<string, []>("softmax_19_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_19_cast_fp16 = softmax(axis = softmax_19_axis_0, x = matmul_19_cast_fp16)[name = tensor<string, []>("softmax_19_cast_fp16")];
            tensor<bool, []> attn_output_77_transpose_x_0 = const()[name = tensor<string, []>("attn_output_77_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_77_transpose_y_0 = const()[name = tensor<string, []>("attn_output_77_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_79_cast_fp16 = transpose(perm = value_states_79_perm_0, x = var_1165_cast_fp16)[name = tensor<string, []>("transpose_193")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_77_cast_fp16 = matmul(transpose_x = attn_output_77_transpose_x_0, transpose_y = attn_output_77_transpose_y_0, x = softmax_19_cast_fp16, y = value_states_79_cast_fp16)[name = tensor<string, []>("attn_output_77_cast_fp16")];
            tensor<int32, [4]> var_1168_perm_0 = const()[name = tensor<string, []>("op_1168_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1170 = const()[name = tensor<string, []>("op_1170"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1168_cast_fp16 = transpose(perm = var_1168_perm_0, x = attn_output_77_cast_fp16)[name = tensor<string, []>("transpose_190")];
            tensor<fp16, [1, 64, 1152]> input_233_cast_fp16 = reshape(shape = var_1170, x = var_1168_cast_fp16)[name = tensor<string, []>("input_233_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(660971520)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(663625792)))];
            tensor<fp16, [1, 64, 1152]> linear_117_cast_fp16 = linear(bias = model_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16, x = input_233_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_235_cast_fp16 = add(x = input_231_cast_fp16, y = linear_117_cast_fp16)[name = tensor<string, []>("input_235_cast_fp16")];
            tensor<int32, [1]> input_237_axes_0 = const()[name = tensor<string, []>("input_237_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(663628160)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(663630528)))];
            tensor<fp16, [1, 64, 1152]> input_237_cast_fp16 = layer_norm(axes = input_237_axes_0, beta = model_text_model_encoder_layers_19_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_19_layer_norm2_weight_to_fp16, x = input_235_cast_fp16)[name = tensor<string, []>("input_237_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(663632896)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673549376)))];
            tensor<fp16, [1, 64, 4304]> linear_118_cast_fp16 = linear(bias = model_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16, x = input_237_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")];
            tensor<string, []> input_241_mode_0 = const()[name = tensor<string, []>("input_241_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_241_cast_fp16 = gelu(mode = input_241_mode_0, x = linear_118_cast_fp16)[name = tensor<string, []>("input_241_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673558080)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683474560)))];
            tensor<fp16, [1, 64, 1152]> linear_119_cast_fp16 = linear(bias = model_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16, x = input_241_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_243_cast_fp16 = add(x = input_235_cast_fp16, y = linear_119_cast_fp16)[name = tensor<string, []>("input_243_cast_fp16")];
            tensor<int32, [1]> hidden_states_121_axes_0 = const()[name = tensor<string, []>("hidden_states_121_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683476928)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683479296)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_121_cast_fp16 = layer_norm(axes = hidden_states_121_axes_0, beta = model_text_model_encoder_layers_20_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_20_layer_norm1_weight_to_fp16, x = input_243_cast_fp16)[name = tensor<string, []>("hidden_states_121_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683481664)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686135936)))];
            tensor<fp16, [1, 64, 1152]> linear_120_cast_fp16 = linear(bias = model_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686138304)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(688792576)))];
            tensor<fp16, [1, 64, 1152]> linear_121_cast_fp16 = linear(bias = model_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(688794944)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691449216)))];
            tensor<fp16, [1, 64, 1152]> linear_122_cast_fp16 = linear(bias = model_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")];
            tensor<int32, [4]> var_1213 = const()[name = tensor<string, []>("op_1213"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1214_cast_fp16 = reshape(shape = var_1213, x = linear_120_cast_fp16)[name = tensor<string, []>("op_1214_cast_fp16")];
            tensor<int32, [4]> var_1216 = const()[name = tensor<string, []>("op_1216"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1217_cast_fp16 = reshape(shape = var_1216, x = linear_121_cast_fp16)[name = tensor<string, []>("op_1217_cast_fp16")];
            tensor<int32, [4]> var_1219 = const()[name = tensor<string, []>("op_1219"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1220_cast_fp16 = reshape(shape = var_1219, x = linear_122_cast_fp16)[name = tensor<string, []>("op_1220_cast_fp16")];
            tensor<int32, [4]> value_states_83_perm_0 = const()[name = tensor<string, []>("value_states_83_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_20_y_0_to_fp16 = const()[name = tensor<string, []>("mul_20_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_20_cast_fp16 = mul(x = var_1214_cast_fp16, y = mul_20_y_0_to_fp16)[name = tensor<string, []>("mul_20_cast_fp16")];
            tensor<bool, []> matmul_20_transpose_y_0 = const()[name = tensor<string, []>("matmul_20_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_20_transpose_x_0 = const()[name = tensor<string, []>("matmul_20_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_148_perm_0 = const()[name = tensor<string, []>("transpose_148_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_149_perm_0 = const()[name = tensor<string, []>("transpose_149_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_149 = transpose(perm = transpose_149_perm_0, x = var_1217_cast_fp16)[name = tensor<string, []>("transpose_187")];
            tensor<fp16, [1, 16, 64, 72]> transpose_148 = transpose(perm = transpose_148_perm_0, x = mul_20_cast_fp16)[name = tensor<string, []>("transpose_188")];
            tensor<fp16, [1, 16, 64, 64]> matmul_20_cast_fp16 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = transpose_148, y = transpose_149)[name = tensor<string, []>("matmul_20_cast_fp16")];
            tensor<int32, []> softmax_20_axis_0 = const()[name = tensor<string, []>("softmax_20_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_20_cast_fp16 = softmax(axis = softmax_20_axis_0, x = matmul_20_cast_fp16)[name = tensor<string, []>("softmax_20_cast_fp16")];
            tensor<bool, []> attn_output_81_transpose_x_0 = const()[name = tensor<string, []>("attn_output_81_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_81_transpose_y_0 = const()[name = tensor<string, []>("attn_output_81_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_83_cast_fp16 = transpose(perm = value_states_83_perm_0, x = var_1220_cast_fp16)[name = tensor<string, []>("transpose_189")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_81_cast_fp16 = matmul(transpose_x = attn_output_81_transpose_x_0, transpose_y = attn_output_81_transpose_y_0, x = softmax_20_cast_fp16, y = value_states_83_cast_fp16)[name = tensor<string, []>("attn_output_81_cast_fp16")];
            tensor<int32, [4]> var_1223_perm_0 = const()[name = tensor<string, []>("op_1223_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1225 = const()[name = tensor<string, []>("op_1225"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1223_cast_fp16 = transpose(perm = var_1223_perm_0, x = attn_output_81_cast_fp16)[name = tensor<string, []>("transpose_186")];
            tensor<fp16, [1, 64, 1152]> input_245_cast_fp16 = reshape(shape = var_1225, x = var_1223_cast_fp16)[name = tensor<string, []>("input_245_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691451584)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(694105856)))];
            tensor<fp16, [1, 64, 1152]> linear_123_cast_fp16 = linear(bias = model_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16, x = input_245_cast_fp16)[name = tensor<string, []>("linear_123_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_247_cast_fp16 = add(x = input_243_cast_fp16, y = linear_123_cast_fp16)[name = tensor<string, []>("input_247_cast_fp16")];
            tensor<int32, [1]> input_249_axes_0 = const()[name = tensor<string, []>("input_249_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(694108224)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(694110592)))];
            tensor<fp16, [1, 64, 1152]> input_249_cast_fp16 = layer_norm(axes = input_249_axes_0, beta = model_text_model_encoder_layers_20_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_20_layer_norm2_weight_to_fp16, x = input_247_cast_fp16)[name = tensor<string, []>("input_249_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(694112960)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(704029440)))];
            tensor<fp16, [1, 64, 4304]> linear_124_cast_fp16 = linear(bias = model_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16, x = input_249_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")];
            tensor<string, []> input_253_mode_0 = const()[name = tensor<string, []>("input_253_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_253_cast_fp16 = gelu(mode = input_253_mode_0, x = linear_124_cast_fp16)[name = tensor<string, []>("input_253_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(704038144)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713954624)))];
            tensor<fp16, [1, 64, 1152]> linear_125_cast_fp16 = linear(bias = model_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16, x = input_253_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_255_cast_fp16 = add(x = input_247_cast_fp16, y = linear_125_cast_fp16)[name = tensor<string, []>("input_255_cast_fp16")];
            tensor<int32, [1]> hidden_states_127_axes_0 = const()[name = tensor<string, []>("hidden_states_127_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713956992)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713959360)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_127_cast_fp16 = layer_norm(axes = hidden_states_127_axes_0, beta = model_text_model_encoder_layers_21_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_21_layer_norm1_weight_to_fp16, x = input_255_cast_fp16)[name = tensor<string, []>("hidden_states_127_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(713961728)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(716616000)))];
            tensor<fp16, [1, 64, 1152]> linear_126_cast_fp16 = linear(bias = model_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(716618368)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(719272640)))];
            tensor<fp16, [1, 64, 1152]> linear_127_cast_fp16 = linear(bias = model_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(719275008)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721929280)))];
            tensor<fp16, [1, 64, 1152]> linear_128_cast_fp16 = linear(bias = model_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")];
            tensor<int32, [4]> var_1268 = const()[name = tensor<string, []>("op_1268"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1269_cast_fp16 = reshape(shape = var_1268, x = linear_126_cast_fp16)[name = tensor<string, []>("op_1269_cast_fp16")];
            tensor<int32, [4]> var_1271 = const()[name = tensor<string, []>("op_1271"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1272_cast_fp16 = reshape(shape = var_1271, x = linear_127_cast_fp16)[name = tensor<string, []>("op_1272_cast_fp16")];
            tensor<int32, [4]> var_1274 = const()[name = tensor<string, []>("op_1274"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1275_cast_fp16 = reshape(shape = var_1274, x = linear_128_cast_fp16)[name = tensor<string, []>("op_1275_cast_fp16")];
            tensor<int32, [4]> value_states_87_perm_0 = const()[name = tensor<string, []>("value_states_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_21_y_0_to_fp16 = const()[name = tensor<string, []>("mul_21_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_21_cast_fp16 = mul(x = var_1269_cast_fp16, y = mul_21_y_0_to_fp16)[name = tensor<string, []>("mul_21_cast_fp16")];
            tensor<bool, []> matmul_21_transpose_y_0 = const()[name = tensor<string, []>("matmul_21_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_21_transpose_x_0 = const()[name = tensor<string, []>("matmul_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_150_perm_0 = const()[name = tensor<string, []>("transpose_150_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_151_perm_0 = const()[name = tensor<string, []>("transpose_151_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_151 = transpose(perm = transpose_151_perm_0, x = var_1272_cast_fp16)[name = tensor<string, []>("transpose_183")];
            tensor<fp16, [1, 16, 64, 72]> transpose_150 = transpose(perm = transpose_150_perm_0, x = mul_21_cast_fp16)[name = tensor<string, []>("transpose_184")];
            tensor<fp16, [1, 16, 64, 64]> matmul_21_cast_fp16 = matmul(transpose_x = matmul_21_transpose_x_0, transpose_y = matmul_21_transpose_y_0, x = transpose_150, y = transpose_151)[name = tensor<string, []>("matmul_21_cast_fp16")];
            tensor<int32, []> softmax_21_axis_0 = const()[name = tensor<string, []>("softmax_21_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_21_cast_fp16 = softmax(axis = softmax_21_axis_0, x = matmul_21_cast_fp16)[name = tensor<string, []>("softmax_21_cast_fp16")];
            tensor<bool, []> attn_output_85_transpose_x_0 = const()[name = tensor<string, []>("attn_output_85_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_85_transpose_y_0 = const()[name = tensor<string, []>("attn_output_85_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_87_cast_fp16 = transpose(perm = value_states_87_perm_0, x = var_1275_cast_fp16)[name = tensor<string, []>("transpose_185")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_85_cast_fp16 = matmul(transpose_x = attn_output_85_transpose_x_0, transpose_y = attn_output_85_transpose_y_0, x = softmax_21_cast_fp16, y = value_states_87_cast_fp16)[name = tensor<string, []>("attn_output_85_cast_fp16")];
            tensor<int32, [4]> var_1278_perm_0 = const()[name = tensor<string, []>("op_1278_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1280 = const()[name = tensor<string, []>("op_1280"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1278_cast_fp16 = transpose(perm = var_1278_perm_0, x = attn_output_85_cast_fp16)[name = tensor<string, []>("transpose_182")];
            tensor<fp16, [1, 64, 1152]> input_257_cast_fp16 = reshape(shape = var_1280, x = var_1278_cast_fp16)[name = tensor<string, []>("input_257_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721931648)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(724585920)))];
            tensor<fp16, [1, 64, 1152]> linear_129_cast_fp16 = linear(bias = model_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16, x = input_257_cast_fp16)[name = tensor<string, []>("linear_129_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_259_cast_fp16 = add(x = input_255_cast_fp16, y = linear_129_cast_fp16)[name = tensor<string, []>("input_259_cast_fp16")];
            tensor<int32, [1]> input_261_axes_0 = const()[name = tensor<string, []>("input_261_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(724588288)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(724590656)))];
            tensor<fp16, [1, 64, 1152]> input_261_cast_fp16 = layer_norm(axes = input_261_axes_0, beta = model_text_model_encoder_layers_21_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_21_layer_norm2_weight_to_fp16, x = input_259_cast_fp16)[name = tensor<string, []>("input_261_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(724593024)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734509504)))];
            tensor<fp16, [1, 64, 4304]> linear_130_cast_fp16 = linear(bias = model_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16, x = input_261_cast_fp16)[name = tensor<string, []>("linear_130_cast_fp16")];
            tensor<string, []> input_265_mode_0 = const()[name = tensor<string, []>("input_265_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_265_cast_fp16 = gelu(mode = input_265_mode_0, x = linear_130_cast_fp16)[name = tensor<string, []>("input_265_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734518208)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(744434688)))];
            tensor<fp16, [1, 64, 1152]> linear_131_cast_fp16 = linear(bias = model_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16, x = input_265_cast_fp16)[name = tensor<string, []>("linear_131_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_267_cast_fp16 = add(x = input_259_cast_fp16, y = linear_131_cast_fp16)[name = tensor<string, []>("input_267_cast_fp16")];
            tensor<int32, [1]> hidden_states_133_axes_0 = const()[name = tensor<string, []>("hidden_states_133_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(744437056)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(744439424)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_133_cast_fp16 = layer_norm(axes = hidden_states_133_axes_0, beta = model_text_model_encoder_layers_22_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_22_layer_norm1_weight_to_fp16, x = input_267_cast_fp16)[name = tensor<string, []>("hidden_states_133_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(744441792)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(747096064)))];
            tensor<fp16, [1, 64, 1152]> linear_132_cast_fp16 = linear(bias = model_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_132_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(747098432)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749752704)))];
            tensor<fp16, [1, 64, 1152]> linear_133_cast_fp16 = linear(bias = model_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749755072)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(752409344)))];
            tensor<fp16, [1, 64, 1152]> linear_134_cast_fp16 = linear(bias = model_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")];
            tensor<int32, [4]> var_1323 = const()[name = tensor<string, []>("op_1323"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1324_cast_fp16 = reshape(shape = var_1323, x = linear_132_cast_fp16)[name = tensor<string, []>("op_1324_cast_fp16")];
            tensor<int32, [4]> var_1326 = const()[name = tensor<string, []>("op_1326"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1327_cast_fp16 = reshape(shape = var_1326, x = linear_133_cast_fp16)[name = tensor<string, []>("op_1327_cast_fp16")];
            tensor<int32, [4]> var_1329 = const()[name = tensor<string, []>("op_1329"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1330_cast_fp16 = reshape(shape = var_1329, x = linear_134_cast_fp16)[name = tensor<string, []>("op_1330_cast_fp16")];
            tensor<int32, [4]> value_states_91_perm_0 = const()[name = tensor<string, []>("value_states_91_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_22_y_0_to_fp16 = const()[name = tensor<string, []>("mul_22_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_22_cast_fp16 = mul(x = var_1324_cast_fp16, y = mul_22_y_0_to_fp16)[name = tensor<string, []>("mul_22_cast_fp16")];
            tensor<bool, []> matmul_22_transpose_y_0 = const()[name = tensor<string, []>("matmul_22_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_22_transpose_x_0 = const()[name = tensor<string, []>("matmul_22_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_152_perm_0 = const()[name = tensor<string, []>("transpose_152_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_153_perm_0 = const()[name = tensor<string, []>("transpose_153_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_153 = transpose(perm = transpose_153_perm_0, x = var_1327_cast_fp16)[name = tensor<string, []>("transpose_179")];
            tensor<fp16, [1, 16, 64, 72]> transpose_152 = transpose(perm = transpose_152_perm_0, x = mul_22_cast_fp16)[name = tensor<string, []>("transpose_180")];
            tensor<fp16, [1, 16, 64, 64]> matmul_22_cast_fp16 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = transpose_152, y = transpose_153)[name = tensor<string, []>("matmul_22_cast_fp16")];
            tensor<int32, []> softmax_22_axis_0 = const()[name = tensor<string, []>("softmax_22_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_22_cast_fp16 = softmax(axis = softmax_22_axis_0, x = matmul_22_cast_fp16)[name = tensor<string, []>("softmax_22_cast_fp16")];
            tensor<bool, []> attn_output_89_transpose_x_0 = const()[name = tensor<string, []>("attn_output_89_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_89_transpose_y_0 = const()[name = tensor<string, []>("attn_output_89_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_91_cast_fp16 = transpose(perm = value_states_91_perm_0, x = var_1330_cast_fp16)[name = tensor<string, []>("transpose_181")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_89_cast_fp16 = matmul(transpose_x = attn_output_89_transpose_x_0, transpose_y = attn_output_89_transpose_y_0, x = softmax_22_cast_fp16, y = value_states_91_cast_fp16)[name = tensor<string, []>("attn_output_89_cast_fp16")];
            tensor<int32, [4]> var_1333_perm_0 = const()[name = tensor<string, []>("op_1333_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1335 = const()[name = tensor<string, []>("op_1335"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1333_cast_fp16 = transpose(perm = var_1333_perm_0, x = attn_output_89_cast_fp16)[name = tensor<string, []>("transpose_178")];
            tensor<fp16, [1, 64, 1152]> input_269_cast_fp16 = reshape(shape = var_1335, x = var_1333_cast_fp16)[name = tensor<string, []>("input_269_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(752411712)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(755065984)))];
            tensor<fp16, [1, 64, 1152]> linear_135_cast_fp16 = linear(bias = model_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16, x = input_269_cast_fp16)[name = tensor<string, []>("linear_135_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_271_cast_fp16 = add(x = input_267_cast_fp16, y = linear_135_cast_fp16)[name = tensor<string, []>("input_271_cast_fp16")];
            tensor<int32, [1]> input_273_axes_0 = const()[name = tensor<string, []>("input_273_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(755068352)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(755070720)))];
            tensor<fp16, [1, 64, 1152]> input_273_cast_fp16 = layer_norm(axes = input_273_axes_0, beta = model_text_model_encoder_layers_22_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_22_layer_norm2_weight_to_fp16, x = input_271_cast_fp16)[name = tensor<string, []>("input_273_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(755073088)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764989568)))];
            tensor<fp16, [1, 64, 4304]> linear_136_cast_fp16 = linear(bias = model_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16, x = input_273_cast_fp16)[name = tensor<string, []>("linear_136_cast_fp16")];
            tensor<string, []> input_277_mode_0 = const()[name = tensor<string, []>("input_277_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_277_cast_fp16 = gelu(mode = input_277_mode_0, x = linear_136_cast_fp16)[name = tensor<string, []>("input_277_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764998272)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(774914752)))];
            tensor<fp16, [1, 64, 1152]> linear_137_cast_fp16 = linear(bias = model_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16, x = input_277_cast_fp16)[name = tensor<string, []>("linear_137_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_279_cast_fp16 = add(x = input_271_cast_fp16, y = linear_137_cast_fp16)[name = tensor<string, []>("input_279_cast_fp16")];
            tensor<int32, [1]> hidden_states_139_axes_0 = const()[name = tensor<string, []>("hidden_states_139_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(774917120)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(774919488)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_139_cast_fp16 = layer_norm(axes = hidden_states_139_axes_0, beta = model_text_model_encoder_layers_23_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_23_layer_norm1_weight_to_fp16, x = input_279_cast_fp16)[name = tensor<string, []>("hidden_states_139_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(774921856)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(777576128)))];
            tensor<fp16, [1, 64, 1152]> linear_138_cast_fp16 = linear(bias = model_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16, x = hidden_states_139_cast_fp16)[name = tensor<string, []>("linear_138_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(777578496)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780232768)))];
            tensor<fp16, [1, 64, 1152]> linear_139_cast_fp16 = linear(bias = model_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16, x = hidden_states_139_cast_fp16)[name = tensor<string, []>("linear_139_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(780235136)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782889408)))];
            tensor<fp16, [1, 64, 1152]> linear_140_cast_fp16 = linear(bias = model_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16, x = hidden_states_139_cast_fp16)[name = tensor<string, []>("linear_140_cast_fp16")];
            tensor<int32, [4]> var_1378 = const()[name = tensor<string, []>("op_1378"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1379_cast_fp16 = reshape(shape = var_1378, x = linear_138_cast_fp16)[name = tensor<string, []>("op_1379_cast_fp16")];
            tensor<int32, [4]> var_1381 = const()[name = tensor<string, []>("op_1381"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1382_cast_fp16 = reshape(shape = var_1381, x = linear_139_cast_fp16)[name = tensor<string, []>("op_1382_cast_fp16")];
            tensor<int32, [4]> var_1384 = const()[name = tensor<string, []>("op_1384"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1385_cast_fp16 = reshape(shape = var_1384, x = linear_140_cast_fp16)[name = tensor<string, []>("op_1385_cast_fp16")];
            tensor<int32, [4]> value_states_95_perm_0 = const()[name = tensor<string, []>("value_states_95_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_23_y_0_to_fp16 = const()[name = tensor<string, []>("mul_23_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_23_cast_fp16 = mul(x = var_1379_cast_fp16, y = mul_23_y_0_to_fp16)[name = tensor<string, []>("mul_23_cast_fp16")];
            tensor<bool, []> matmul_23_transpose_y_0 = const()[name = tensor<string, []>("matmul_23_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_23_transpose_x_0 = const()[name = tensor<string, []>("matmul_23_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_154_perm_0 = const()[name = tensor<string, []>("transpose_154_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_155_perm_0 = const()[name = tensor<string, []>("transpose_155_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_155 = transpose(perm = transpose_155_perm_0, x = var_1382_cast_fp16)[name = tensor<string, []>("transpose_175")];
            tensor<fp16, [1, 16, 64, 72]> transpose_154 = transpose(perm = transpose_154_perm_0, x = mul_23_cast_fp16)[name = tensor<string, []>("transpose_176")];
            tensor<fp16, [1, 16, 64, 64]> matmul_23_cast_fp16 = matmul(transpose_x = matmul_23_transpose_x_0, transpose_y = matmul_23_transpose_y_0, x = transpose_154, y = transpose_155)[name = tensor<string, []>("matmul_23_cast_fp16")];
            tensor<int32, []> softmax_23_axis_0 = const()[name = tensor<string, []>("softmax_23_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_23_cast_fp16 = softmax(axis = softmax_23_axis_0, x = matmul_23_cast_fp16)[name = tensor<string, []>("softmax_23_cast_fp16")];
            tensor<bool, []> attn_output_93_transpose_x_0 = const()[name = tensor<string, []>("attn_output_93_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_93_transpose_y_0 = const()[name = tensor<string, []>("attn_output_93_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_95_cast_fp16 = transpose(perm = value_states_95_perm_0, x = var_1385_cast_fp16)[name = tensor<string, []>("transpose_177")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_93_cast_fp16 = matmul(transpose_x = attn_output_93_transpose_x_0, transpose_y = attn_output_93_transpose_y_0, x = softmax_23_cast_fp16, y = value_states_95_cast_fp16)[name = tensor<string, []>("attn_output_93_cast_fp16")];
            tensor<int32, [4]> var_1388_perm_0 = const()[name = tensor<string, []>("op_1388_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1390 = const()[name = tensor<string, []>("op_1390"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1388_cast_fp16 = transpose(perm = var_1388_perm_0, x = attn_output_93_cast_fp16)[name = tensor<string, []>("transpose_174")];
            tensor<fp16, [1, 64, 1152]> input_281_cast_fp16 = reshape(shape = var_1390, x = var_1388_cast_fp16)[name = tensor<string, []>("input_281_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782891776)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785546048)))];
            tensor<fp16, [1, 64, 1152]> linear_141_cast_fp16 = linear(bias = model_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16, x = input_281_cast_fp16)[name = tensor<string, []>("linear_141_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_283_cast_fp16 = add(x = input_279_cast_fp16, y = linear_141_cast_fp16)[name = tensor<string, []>("input_283_cast_fp16")];
            tensor<int32, [1]> input_285_axes_0 = const()[name = tensor<string, []>("input_285_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785548416)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785550784)))];
            tensor<fp16, [1, 64, 1152]> input_285_cast_fp16 = layer_norm(axes = input_285_axes_0, beta = model_text_model_encoder_layers_23_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_23_layer_norm2_weight_to_fp16, x = input_283_cast_fp16)[name = tensor<string, []>("input_285_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(785553152)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795469632)))];
            tensor<fp16, [1, 64, 4304]> linear_142_cast_fp16 = linear(bias = model_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16, x = input_285_cast_fp16)[name = tensor<string, []>("linear_142_cast_fp16")];
            tensor<string, []> input_289_mode_0 = const()[name = tensor<string, []>("input_289_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_289_cast_fp16 = gelu(mode = input_289_mode_0, x = linear_142_cast_fp16)[name = tensor<string, []>("input_289_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795478336)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805394816)))];
            tensor<fp16, [1, 64, 1152]> linear_143_cast_fp16 = linear(bias = model_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16, x = input_289_cast_fp16)[name = tensor<string, []>("linear_143_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_291_cast_fp16 = add(x = input_283_cast_fp16, y = linear_143_cast_fp16)[name = tensor<string, []>("input_291_cast_fp16")];
            tensor<int32, [1]> hidden_states_145_axes_0 = const()[name = tensor<string, []>("hidden_states_145_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805397184)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805399552)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_145_cast_fp16 = layer_norm(axes = hidden_states_145_axes_0, beta = model_text_model_encoder_layers_24_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_24_layer_norm1_weight_to_fp16, x = input_291_cast_fp16)[name = tensor<string, []>("hidden_states_145_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805401920)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808056192)))];
            tensor<fp16, [1, 64, 1152]> linear_144_cast_fp16 = linear(bias = model_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16, x = hidden_states_145_cast_fp16)[name = tensor<string, []>("linear_144_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808058560)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810712832)))];
            tensor<fp16, [1, 64, 1152]> linear_145_cast_fp16 = linear(bias = model_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16, x = hidden_states_145_cast_fp16)[name = tensor<string, []>("linear_145_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810715200)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(813369472)))];
            tensor<fp16, [1, 64, 1152]> linear_146_cast_fp16 = linear(bias = model_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16, x = hidden_states_145_cast_fp16)[name = tensor<string, []>("linear_146_cast_fp16")];
            tensor<int32, [4]> var_1433 = const()[name = tensor<string, []>("op_1433"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1434_cast_fp16 = reshape(shape = var_1433, x = linear_144_cast_fp16)[name = tensor<string, []>("op_1434_cast_fp16")];
            tensor<int32, [4]> var_1436 = const()[name = tensor<string, []>("op_1436"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1437_cast_fp16 = reshape(shape = var_1436, x = linear_145_cast_fp16)[name = tensor<string, []>("op_1437_cast_fp16")];
            tensor<int32, [4]> var_1439 = const()[name = tensor<string, []>("op_1439"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1440_cast_fp16 = reshape(shape = var_1439, x = linear_146_cast_fp16)[name = tensor<string, []>("op_1440_cast_fp16")];
            tensor<int32, [4]> value_states_99_perm_0 = const()[name = tensor<string, []>("value_states_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_24_y_0_to_fp16 = const()[name = tensor<string, []>("mul_24_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_24_cast_fp16 = mul(x = var_1434_cast_fp16, y = mul_24_y_0_to_fp16)[name = tensor<string, []>("mul_24_cast_fp16")];
            tensor<bool, []> matmul_24_transpose_y_0 = const()[name = tensor<string, []>("matmul_24_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_24_transpose_x_0 = const()[name = tensor<string, []>("matmul_24_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_156_perm_0 = const()[name = tensor<string, []>("transpose_156_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_157_perm_0 = const()[name = tensor<string, []>("transpose_157_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_157 = transpose(perm = transpose_157_perm_0, x = var_1437_cast_fp16)[name = tensor<string, []>("transpose_171")];
            tensor<fp16, [1, 16, 64, 72]> transpose_156 = transpose(perm = transpose_156_perm_0, x = mul_24_cast_fp16)[name = tensor<string, []>("transpose_172")];
            tensor<fp16, [1, 16, 64, 64]> matmul_24_cast_fp16 = matmul(transpose_x = matmul_24_transpose_x_0, transpose_y = matmul_24_transpose_y_0, x = transpose_156, y = transpose_157)[name = tensor<string, []>("matmul_24_cast_fp16")];
            tensor<int32, []> softmax_24_axis_0 = const()[name = tensor<string, []>("softmax_24_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_24_cast_fp16 = softmax(axis = softmax_24_axis_0, x = matmul_24_cast_fp16)[name = tensor<string, []>("softmax_24_cast_fp16")];
            tensor<bool, []> attn_output_97_transpose_x_0 = const()[name = tensor<string, []>("attn_output_97_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_97_transpose_y_0 = const()[name = tensor<string, []>("attn_output_97_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_99_cast_fp16 = transpose(perm = value_states_99_perm_0, x = var_1440_cast_fp16)[name = tensor<string, []>("transpose_173")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_97_cast_fp16 = matmul(transpose_x = attn_output_97_transpose_x_0, transpose_y = attn_output_97_transpose_y_0, x = softmax_24_cast_fp16, y = value_states_99_cast_fp16)[name = tensor<string, []>("attn_output_97_cast_fp16")];
            tensor<int32, [4]> var_1443_perm_0 = const()[name = tensor<string, []>("op_1443_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1445 = const()[name = tensor<string, []>("op_1445"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1443_cast_fp16 = transpose(perm = var_1443_perm_0, x = attn_output_97_cast_fp16)[name = tensor<string, []>("transpose_170")];
            tensor<fp16, [1, 64, 1152]> input_293_cast_fp16 = reshape(shape = var_1445, x = var_1443_cast_fp16)[name = tensor<string, []>("input_293_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(813371840)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(816026112)))];
            tensor<fp16, [1, 64, 1152]> linear_147_cast_fp16 = linear(bias = model_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16, x = input_293_cast_fp16)[name = tensor<string, []>("linear_147_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_295_cast_fp16 = add(x = input_291_cast_fp16, y = linear_147_cast_fp16)[name = tensor<string, []>("input_295_cast_fp16")];
            tensor<int32, [1]> input_297_axes_0 = const()[name = tensor<string, []>("input_297_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(816028480)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(816030848)))];
            tensor<fp16, [1, 64, 1152]> input_297_cast_fp16 = layer_norm(axes = input_297_axes_0, beta = model_text_model_encoder_layers_24_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_24_layer_norm2_weight_to_fp16, x = input_295_cast_fp16)[name = tensor<string, []>("input_297_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(816033216)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(825949696)))];
            tensor<fp16, [1, 64, 4304]> linear_148_cast_fp16 = linear(bias = model_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16, x = input_297_cast_fp16)[name = tensor<string, []>("linear_148_cast_fp16")];
            tensor<string, []> input_301_mode_0 = const()[name = tensor<string, []>("input_301_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_301_cast_fp16 = gelu(mode = input_301_mode_0, x = linear_148_cast_fp16)[name = tensor<string, []>("input_301_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(825958400)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835874880)))];
            tensor<fp16, [1, 64, 1152]> linear_149_cast_fp16 = linear(bias = model_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16, x = input_301_cast_fp16)[name = tensor<string, []>("linear_149_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_303_cast_fp16 = add(x = input_295_cast_fp16, y = linear_149_cast_fp16)[name = tensor<string, []>("input_303_cast_fp16")];
            tensor<int32, [1]> hidden_states_151_axes_0 = const()[name = tensor<string, []>("hidden_states_151_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835877248)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835879616)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_151_cast_fp16 = layer_norm(axes = hidden_states_151_axes_0, beta = model_text_model_encoder_layers_25_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_25_layer_norm1_weight_to_fp16, x = input_303_cast_fp16)[name = tensor<string, []>("hidden_states_151_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835881984)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838536256)))];
            tensor<fp16, [1, 64, 1152]> linear_150_cast_fp16 = linear(bias = model_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16, x = hidden_states_151_cast_fp16)[name = tensor<string, []>("linear_150_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838538624)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841192896)))];
            tensor<fp16, [1, 64, 1152]> linear_151_cast_fp16 = linear(bias = model_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16, x = hidden_states_151_cast_fp16)[name = tensor<string, []>("linear_151_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841195264)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(843849536)))];
            tensor<fp16, [1, 64, 1152]> linear_152_cast_fp16 = linear(bias = model_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16, x = hidden_states_151_cast_fp16)[name = tensor<string, []>("linear_152_cast_fp16")];
            tensor<int32, [4]> var_1488 = const()[name = tensor<string, []>("op_1488"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1489_cast_fp16 = reshape(shape = var_1488, x = linear_150_cast_fp16)[name = tensor<string, []>("op_1489_cast_fp16")];
            tensor<int32, [4]> var_1491 = const()[name = tensor<string, []>("op_1491"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1492_cast_fp16 = reshape(shape = var_1491, x = linear_151_cast_fp16)[name = tensor<string, []>("op_1492_cast_fp16")];
            tensor<int32, [4]> var_1494 = const()[name = tensor<string, []>("op_1494"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1495_cast_fp16 = reshape(shape = var_1494, x = linear_152_cast_fp16)[name = tensor<string, []>("op_1495_cast_fp16")];
            tensor<int32, [4]> value_states_103_perm_0 = const()[name = tensor<string, []>("value_states_103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_25_y_0_to_fp16 = const()[name = tensor<string, []>("mul_25_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_25_cast_fp16 = mul(x = var_1489_cast_fp16, y = mul_25_y_0_to_fp16)[name = tensor<string, []>("mul_25_cast_fp16")];
            tensor<bool, []> matmul_25_transpose_y_0 = const()[name = tensor<string, []>("matmul_25_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_25_transpose_x_0 = const()[name = tensor<string, []>("matmul_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_158_perm_0 = const()[name = tensor<string, []>("transpose_158_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_159_perm_0 = const()[name = tensor<string, []>("transpose_159_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_159 = transpose(perm = transpose_159_perm_0, x = var_1492_cast_fp16)[name = tensor<string, []>("transpose_167")];
            tensor<fp16, [1, 16, 64, 72]> transpose_158 = transpose(perm = transpose_158_perm_0, x = mul_25_cast_fp16)[name = tensor<string, []>("transpose_168")];
            tensor<fp16, [1, 16, 64, 64]> matmul_25_cast_fp16 = matmul(transpose_x = matmul_25_transpose_x_0, transpose_y = matmul_25_transpose_y_0, x = transpose_158, y = transpose_159)[name = tensor<string, []>("matmul_25_cast_fp16")];
            tensor<int32, []> softmax_25_axis_0 = const()[name = tensor<string, []>("softmax_25_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_25_cast_fp16 = softmax(axis = softmax_25_axis_0, x = matmul_25_cast_fp16)[name = tensor<string, []>("softmax_25_cast_fp16")];
            tensor<bool, []> attn_output_101_transpose_x_0 = const()[name = tensor<string, []>("attn_output_101_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_101_transpose_y_0 = const()[name = tensor<string, []>("attn_output_101_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_103_cast_fp16 = transpose(perm = value_states_103_perm_0, x = var_1495_cast_fp16)[name = tensor<string, []>("transpose_169")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_101_cast_fp16 = matmul(transpose_x = attn_output_101_transpose_x_0, transpose_y = attn_output_101_transpose_y_0, x = softmax_25_cast_fp16, y = value_states_103_cast_fp16)[name = tensor<string, []>("attn_output_101_cast_fp16")];
            tensor<int32, [4]> var_1498_perm_0 = const()[name = tensor<string, []>("op_1498_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1500 = const()[name = tensor<string, []>("op_1500"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1498_cast_fp16 = transpose(perm = var_1498_perm_0, x = attn_output_101_cast_fp16)[name = tensor<string, []>("transpose_166")];
            tensor<fp16, [1, 64, 1152]> input_305_cast_fp16 = reshape(shape = var_1500, x = var_1498_cast_fp16)[name = tensor<string, []>("input_305_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(843851904)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(846506176)))];
            tensor<fp16, [1, 64, 1152]> linear_153_cast_fp16 = linear(bias = model_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16, x = input_305_cast_fp16)[name = tensor<string, []>("linear_153_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_307_cast_fp16 = add(x = input_303_cast_fp16, y = linear_153_cast_fp16)[name = tensor<string, []>("input_307_cast_fp16")];
            tensor<int32, [1]> input_309_axes_0 = const()[name = tensor<string, []>("input_309_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(846508544)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(846510912)))];
            tensor<fp16, [1, 64, 1152]> input_309_cast_fp16 = layer_norm(axes = input_309_axes_0, beta = model_text_model_encoder_layers_25_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_25_layer_norm2_weight_to_fp16, x = input_307_cast_fp16)[name = tensor<string, []>("input_309_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(846513280)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(856429760)))];
            tensor<fp16, [1, 64, 4304]> linear_154_cast_fp16 = linear(bias = model_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16, x = input_309_cast_fp16)[name = tensor<string, []>("linear_154_cast_fp16")];
            tensor<string, []> input_313_mode_0 = const()[name = tensor<string, []>("input_313_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_313_cast_fp16 = gelu(mode = input_313_mode_0, x = linear_154_cast_fp16)[name = tensor<string, []>("input_313_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(856438464)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(866354944)))];
            tensor<fp16, [1, 64, 1152]> linear_155_cast_fp16 = linear(bias = model_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16, x = input_313_cast_fp16)[name = tensor<string, []>("linear_155_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_315_cast_fp16 = add(x = input_307_cast_fp16, y = linear_155_cast_fp16)[name = tensor<string, []>("input_315_cast_fp16")];
            tensor<int32, [1]> hidden_states_157_axes_0 = const()[name = tensor<string, []>("hidden_states_157_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(866357312)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(866359680)))];
            tensor<fp16, [1, 64, 1152]> hidden_states_157_cast_fp16 = layer_norm(axes = hidden_states_157_axes_0, beta = model_text_model_encoder_layers_26_layer_norm1_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_26_layer_norm1_weight_to_fp16, x = input_315_cast_fp16)[name = tensor<string, []>("hidden_states_157_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(866362048)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(869016320)))];
            tensor<fp16, [1, 64, 1152]> linear_156_cast_fp16 = linear(bias = model_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16, weight = model_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16, x = hidden_states_157_cast_fp16)[name = tensor<string, []>("linear_156_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(869018688)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(871672960)))];
            tensor<fp16, [1, 64, 1152]> linear_157_cast_fp16 = linear(bias = model_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16, weight = model_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16, x = hidden_states_157_cast_fp16)[name = tensor<string, []>("linear_157_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(871675328)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(874329600)))];
            tensor<fp16, [1, 64, 1152]> linear_158_cast_fp16 = linear(bias = model_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16, weight = model_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16, x = hidden_states_157_cast_fp16)[name = tensor<string, []>("linear_158_cast_fp16")];
            tensor<int32, [4]> var_1543 = const()[name = tensor<string, []>("op_1543"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1544_cast_fp16 = reshape(shape = var_1543, x = linear_156_cast_fp16)[name = tensor<string, []>("op_1544_cast_fp16")];
            tensor<int32, [4]> var_1546 = const()[name = tensor<string, []>("op_1546"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1547_cast_fp16 = reshape(shape = var_1546, x = linear_157_cast_fp16)[name = tensor<string, []>("op_1547_cast_fp16")];
            tensor<int32, [4]> var_1549 = const()[name = tensor<string, []>("op_1549"), val = tensor<int32, [4]>([1, 64, 16, 72])];
            tensor<fp16, [1, 64, 16, 72]> var_1550_cast_fp16 = reshape(shape = var_1549, x = linear_158_cast_fp16)[name = tensor<string, []>("op_1550_cast_fp16")];
            tensor<int32, [4]> value_states_perm_0 = const()[name = tensor<string, []>("value_states_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, []> mul_26_y_0_to_fp16 = const()[name = tensor<string, []>("mul_26_y_0_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 64, 16, 72]> mul_26_cast_fp16 = mul(x = var_1544_cast_fp16, y = mul_26_y_0_to_fp16)[name = tensor<string, []>("mul_26_cast_fp16")];
            tensor<bool, []> matmul_26_transpose_y_0 = const()[name = tensor<string, []>("matmul_26_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_26_transpose_x_0 = const()[name = tensor<string, []>("matmul_26_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_160_perm_0 = const()[name = tensor<string, []>("transpose_160_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_161_perm_0 = const()[name = tensor<string, []>("transpose_161_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 64, 72]> transpose_161 = transpose(perm = transpose_161_perm_0, x = var_1547_cast_fp16)[name = tensor<string, []>("transpose_163")];
            tensor<fp16, [1, 16, 64, 72]> transpose_160 = transpose(perm = transpose_160_perm_0, x = mul_26_cast_fp16)[name = tensor<string, []>("transpose_164")];
            tensor<fp16, [1, 16, 64, 64]> matmul_26_cast_fp16 = matmul(transpose_x = matmul_26_transpose_x_0, transpose_y = matmul_26_transpose_y_0, x = transpose_160, y = transpose_161)[name = tensor<string, []>("matmul_26_cast_fp16")];
            tensor<int32, []> softmax_26_axis_0 = const()[name = tensor<string, []>("softmax_26_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 64, 64]> softmax_26_cast_fp16 = softmax(axis = softmax_26_axis_0, x = matmul_26_cast_fp16)[name = tensor<string, []>("softmax_26_cast_fp16")];
            tensor<bool, []> attn_output_105_transpose_x_0 = const()[name = tensor<string, []>("attn_output_105_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_105_transpose_y_0 = const()[name = tensor<string, []>("attn_output_105_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 64, 72]> value_states_cast_fp16 = transpose(perm = value_states_perm_0, x = var_1550_cast_fp16)[name = tensor<string, []>("transpose_165")];
            tensor<fp16, [1, 16, 64, 72]> attn_output_105_cast_fp16 = matmul(transpose_x = attn_output_105_transpose_x_0, transpose_y = attn_output_105_transpose_y_0, x = softmax_26_cast_fp16, y = value_states_cast_fp16)[name = tensor<string, []>("attn_output_105_cast_fp16")];
            tensor<int32, [4]> var_1553_perm_0 = const()[name = tensor<string, []>("op_1553_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1555 = const()[name = tensor<string, []>("op_1555"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<fp16, [1, 64, 16, 72]> var_1553_cast_fp16 = transpose(perm = var_1553_perm_0, x = attn_output_105_cast_fp16)[name = tensor<string, []>("transpose_162")];
            tensor<fp16, [1, 64, 1152]> input_317_cast_fp16 = reshape(shape = var_1555, x = var_1553_cast_fp16)[name = tensor<string, []>("input_317_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(874331968)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(876986240)))];
            tensor<fp16, [1, 64, 1152]> linear_159_cast_fp16 = linear(bias = model_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16, weight = model_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16, x = input_317_cast_fp16)[name = tensor<string, []>("linear_159_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_319_cast_fp16 = add(x = input_315_cast_fp16, y = linear_159_cast_fp16)[name = tensor<string, []>("input_319_cast_fp16")];
            tensor<int32, [1]> input_321_axes_0 = const()[name = tensor<string, []>("input_321_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(876988608)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(876990976)))];
            tensor<fp16, [1, 64, 1152]> input_321_cast_fp16 = layer_norm(axes = input_321_axes_0, beta = model_text_model_encoder_layers_26_layer_norm2_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_encoder_layers_26_layer_norm2_weight_to_fp16, x = input_319_cast_fp16)[name = tensor<string, []>("input_321_cast_fp16")];
            tensor<fp16, [4304, 1152]> model_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(876993344)))];
            tensor<fp16, [4304]> model_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886909824)))];
            tensor<fp16, [1, 64, 4304]> linear_160_cast_fp16 = linear(bias = model_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16, weight = model_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16, x = input_321_cast_fp16)[name = tensor<string, []>("linear_160_cast_fp16")];
            tensor<string, []> input_325_mode_0 = const()[name = tensor<string, []>("input_325_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 64, 4304]> input_325_cast_fp16 = gelu(mode = input_325_mode_0, x = linear_160_cast_fp16)[name = tensor<string, []>("input_325_cast_fp16")];
            tensor<fp16, [1152, 4304]> model_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886918528)))];
            tensor<fp16, [1152]> model_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896835008)))];
            tensor<fp16, [1, 64, 1152]> linear_161_cast_fp16 = linear(bias = model_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16, weight = model_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16, x = input_325_cast_fp16)[name = tensor<string, []>("linear_161_cast_fp16")];
            tensor<fp16, [1, 64, 1152]> input_327_cast_fp16 = add(x = input_319_cast_fp16, y = linear_161_cast_fp16)[name = tensor<string, []>("input_327_cast_fp16")];
            tensor<int32, [1]> last_hidden_state_axes_0 = const()[name = tensor<string, []>("last_hidden_state_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> model_text_model_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896837376)))];
            tensor<fp16, [1152]> model_text_model_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896839744)))];
            tensor<fp16, [1, 64, 1152]> last_hidden_state_cast_fp16 = layer_norm(axes = last_hidden_state_axes_0, beta = model_text_model_final_layer_norm_bias_to_fp16, epsilon = var_11_to_fp16, gamma = model_text_model_final_layer_norm_weight_to_fp16, x = input_327_cast_fp16)[name = tensor<string, []>("last_hidden_state_cast_fp16")];
            tensor<int32, [3]> var_1580_begin_0 = const()[name = tensor<string, []>("op_1580_begin_0"), val = tensor<int32, [3]>([0, -1, 0])];
            tensor<int32, [3]> var_1580_end_0 = const()[name = tensor<string, []>("op_1580_end_0"), val = tensor<int32, [3]>([1, 64, 1152])];
            tensor<bool, [3]> var_1580_end_mask_0 = const()[name = tensor<string, []>("op_1580_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<bool, [3]> var_1580_squeeze_mask_0 = const()[name = tensor<string, []>("op_1580_squeeze_mask_0"), val = tensor<bool, [3]>([false, true, false])];
            tensor<fp16, [1, 1152]> var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, squeeze_mask = var_1580_squeeze_mask_0, x = last_hidden_state_cast_fp16)[name = tensor<string, []>("op_1580_cast_fp16")];
            tensor<fp16, [1152, 1152]> model_text_model_head_weight_to_fp16 = const()[name = tensor<string, []>("model_text_model_head_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(896842112)))];
            tensor<fp16, [1152]> model_text_model_head_bias_to_fp16 = const()[name = tensor<string, []>("model_text_model_head_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(899496384)))];
            tensor<fp16, [1, 1152]> linear_162_cast_fp16 = linear(bias = model_text_model_head_bias_to_fp16, weight = model_text_model_head_weight_to_fp16, x = var_1580_cast_fp16)[name = tensor<string, []>("linear_162_cast_fp16")];
            tensor<int32, [1]> var_1587 = const()[name = tensor<string, []>("op_1587"), val = tensor<int32, [1]>([-1])];
            tensor<bool, []> var_1588 = const()[name = tensor<string, []>("op_1588"), val = tensor<bool, []>(true)];
            tensor<fp16, [1, 1]> var_1590_cast_fp16 = reduce_l2_norm(axes = var_1587, keep_dims = var_1588, x = linear_162_cast_fp16)[name = tensor<string, []>("op_1590_cast_fp16")];
            tensor<fp16, [1, 1152]> embedding = real_div(x = linear_162_cast_fp16, y = var_1590_cast_fp16)[name = tensor<string, []>("op_1591_cast_fp16")];
        } -> (embedding);
}