OpBuilder builder(module);

  std::vector<Type> input_types = GetValueTypes(inputs);
  std::vector<Type> output_types = GetValueTypes(outputs);
  builder.setInsertionPointToEnd(&module.getBodyRegion().back());
  func::FuncOp func_op = builder.create<func::FuncOp>(
      module.getLoc(), name,
      builder.getFunctionType(input_types, output_types));
  func_op.setPrivate();

  return func_op;
}

  } else {
    // Graphdef import path.
    module = tensorflow::LoadFromGraphdefOrMlirSource(
        input_file_name, input_mlir, use_splatted_constant, custom_opdefs,
        specs, debug_info_file, input_arrays, input_dtypes, input_shapes,
        output_arrays, control_output_arrays, &source_mgr, &context);
  }

  // If errors occur, the library call in the above already logged the error
  // message. So we can just return here.

  auto* cc_ctx = reinterpret_cast<::tensorflow::OpKernelContext*>(ctx);
  CHECK_GE(index, 0);                     // Crash OK
  CHECK_LT(index, cc_ctx->num_inputs());  // Crash OK
  return static_cast<TF_DataType>(cc_ctx->input_dtype(index));
}

void TF_SetOutput(TF_OpKernelContext* ctx, int i, const TF_Tensor* tensor,
                  TF_Status* status) {
  auto* cc_ctx = reinterpret_cast<::tensorflow::OpKernelContext*>(ctx);