}
        bool found = false;
        for (int j = 0; j < op_it->second.output_tensor_info.size(); ++j) {
          if (op_it->second.output_tensor_info[j].GetID() == id) {
            found = true;
            Gradient* ones_like = nullptr;
            TF_RETURN_IF_ERROR(vspace.BuildOnesLike(
                op_it->second.output_tensor_info[j], &ones_like));
            (*result)[id].push_back(ones_like);
            break;

// tensors will included in the input properties.
// If include_output_tensor_values is true, the values of constant tensors will
// be included in the output properties.
TF_CAPI_EXPORT extern void TF_InferStatically(
    TF_GraphProperties* graph_properties, TF_Bool assume_valid_feeds,
    TF_Bool aggressive_shape_inference, TF_Bool include_input_tensor_values,
    TF_Bool include_output_tensor_values, TF_Status* s);

                        TF_Bool assume_valid_feeds,
                        TF_Bool aggressive_shape_inference,
                        TF_Bool include_input_tensor_values,
                        TF_Bool include_output_tensor_values,
                        TF_Status* status) {
  TF_SetStatus(status, TF_OK, "");
  absl::Status s =
      reinterpret_cast<tensorflow::grappler::GraphProperties*>(graph_properties)

  void SetOutputs(const std::vector<TF_Output>& outputs);
  void SetTargets(std::initializer_list<TF_Operation*> targets);

  void Run(TF_Status* s);

  void CloseAndDelete(TF_Status* s);

  TF_Tensor* output_tensor(int i) { return output_values_[i]; }

  TF_Session* mutable_session() { return session_; }

 private:
  void DeleteInputValues();
  void ResetOutputValues();

  TF_Session* session_;

    csession->SetOutputs(grad_outputs_vec);
    csession->Run(s_);
    ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
    TF_Tensor* out0 = csession->output_tensor(0);
    TF_Tensor* out1 = csession->output_tensor(1);

    std::vector<TF_Output> expected_grad_outputs_vec;
    expected_grad_outputs_vec.assign(expected_grad_outputs,
                                     expected_grad_outputs + 2);

    const std::vector<std::pair<string, Tensor>>& input_pairs,
    // Output tensors
    const std::vector<string>& output_tensor_names, TF_Tensor** c_outputs,
    // Target nodes
    const std::vector<string>& target_oper_names, TF_Buffer* run_metadata,
    TF_Status* status) {
  const int noutputs = output_tensor_names.size();
  std::vector<Tensor> outputs(noutputs);
  Status result;

  if (handle == nullptr) {

    csession.Run(s_);
    ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);

    // Check results
    for (int i = 0; i < expected_results.size(); ++i) {
      TF_Tensor* out = csession.output_tensor(i);
      ASSERT_TRUE(out != nullptr);
      EXPECT_EQ(TF_INT32, TF_TensorType(out));
      EXPECT_EQ(1, TF_NumDims(out));
      CompareInt32Tensor(expected_results[i], out);
    }
  }

  std::vector<tensorflow::OutputTensor> input_tensors;
  std::unordered_map<const Node*, std::vector<int>> input_nodes;
  status->status = tensorflow::ProcessInputs(fn_body, fn_name, ninputs, inputs,
                                             &input_tensors, &input_nodes);
  if (TF_GetCode(status) != TF_OK) return nullptr;

  // Process outputs.
  std::vector<tensorflow::OutputTensor> output_tensors;

       *   `tflite::Interpreter::signature_inputs`
       *   `tflite::Interpreter::signature_outputs`
       *   `tflite::Interpreter::input_tensor_by_signature`
       *   `tflite::Interpreter::output_tensor_by_signature`

    *  Similarly, the following signature runner functions in the TF Lite C API are no longer considered experimental:

       *    `TfLiteInterpreterGetSignatureCount`