// within the subgraph and are referenced outside the subgraph.
  llvm::SmallVector<Type> input_types =
      TypesFromValues(subgraph.FuncArguments());
  llvm::SmallVector<Type> return_types =
      TypesFromValues(subgraph.FuncOutputs());

  FunctionType function_type =
      builder.getFunctionType(input_types, return_types);

  std::string function_name = absl::StrCat("func_", subgraph.subgraph_id_);

    auto* context = shape_refiner_->GetContext(&node);
    DataType dtype = node.input_type(idx);
    return ConvertDataTypeAndShape(dtype, context->input(idx),
                                   context->input_handle_shapes_and_types(idx),
                                   context, builder);
  }
  DataType dtype = node.properties()->input_types[idx];
  mlir::Type element_type;

  mlir::MLIRContext* context = tfr_module_.getContext();
  llvm::SmallVector<mlir::Type, 4> input_tys, output_tys;
  mlir::Builder builder(context);
  for (auto ty : input_dtys) {
    mlir::Type elt_ty;
    TF_RETURN_IF_ERROR(ConvertDataType(ty, builder, &elt_ty));
    mlir::TensorType mlir_ty = mlir::UnrankedTensorType::get(elt_ty);
    input_tys.push_back(mlir_ty);
  }
  for (auto ty : output_dtys) {
    mlir::Type elt_ty;

// attaching a to the initializer function's type so that it is identifiable.
Location CreateInitOpLoc(MLIRContext* ctx, func::FuncOp init_func_ops) {
  const std::string init_type = GetInitializerType(init_func_ops);
  const std::string name =
      absl::StrCat(init_type, "_", init_func_ops.getName().str());
  return NameLoc::get(StringAttr::get(ctx, name));
}

void MergeInitializerFunctionOpsToMainPass::runOnOperation() {

      enable_shape_inference, unconditionally_use_set_output_shapes,
      enable_soft_placement,  set_original_tf_func_name};

  auto module_or = tensorflow::GraphdefToMlirTranslateFunction(
      input, input_arrays, input_dtypes, input_shapes, output_arrays,
      control_output_arrays, options, context);
  if (!module_or.status().ok()) return nullptr;
  return std::move(module_or).value();
}

  if (!s.ok()) {
    return std::nullopt;
  }
  return attr_value->s();
}

bool HasResourceInputOrOutput(const Node& node) {
  return std::find(node.input_types().begin(), node.input_types().end(),
                   DT_RESOURCE) != node.input_types().end() ||
         std::find(node.output_types().begin(), node.output_types().end(),
                   DT_RESOURCE) != node.output_types().end();
}

absl::StatusOr<ElementsAttr> ConvertTensor(const Tensor& input_tensor,
                                           Builder* builder) {
  const auto& input_dtype = input_tensor.dtype();
  const auto& input_shape = input_tensor.shape();
  Type elt_type;
  TF_RETURN_IF_ERROR(ConvertDataType(input_dtype, *builder, &elt_type));
  SmallVector<int64_t, 4> shape;
  ConvertToMlirShape(input_shape, &shape);

    absl::Span<const int64_t> input_tensor_id,
    absl::Span<const tensorflow::DataType> input_dtypes,
    const std::function<BackwardFunction*()>& backward_function_getter,
    const std::function<void(BackwardFunction*)>& backward_function_deleter) {
  if (!ShouldRecord(input_tensor_id, input_dtypes)) {
    return;
  }
  std::vector<int64_t> ids;
  ids.reserve(input_tensor_id.size());

  // Populate inputs.
  std::vector<DataType> input_dtypes;
  TF_RETURN_IF_ERROR(GetNodeAttr(call_node->attrs(), "Tinputs", &input_dtypes));
  std::vector<NodeDefBuilder::NodeOut> inputs(input_dtypes.size());
  for (auto e : call_node->in_edges()) {
    if (e->IsControlEdge()) {
      continue;
    }

    const int input_dtypes_size = input_dtypes.size();

                           const string& op_name) {
  std::vector<int64_t> input_ids(inputs.size());
  std::vector<tensorflow::DataType> input_dtypes(inputs.size());
  for (int i = 0; i < inputs.size(); i++) {
    input_ids[i] = ToId(inputs[i]);
    input_dtypes[i] = inputs[i]->DataType();
  }
  std::vector<TapeTensor> tape_tensors;
  tape_tensors.reserve(outputs.size());
  for (auto t : outputs) {