public:
  explicit ReluGradientFunction(vector<AbstractTensorHandle*> f_outputs)
      : forward_outputs_(f_outputs) {
    for (auto output : forward_outputs_) {
      if (output) {
        output->Ref();
      }
    }
  }

  Status Compute(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {

  }
  return outputs;
}

void TFOutputsFromOutputs(const std::vector<tensorflow::Output>& outputs,
                          TF_Output* tf_outputs) {
  for (int i = 0; i < outputs.size(); i++) {
    tf_outputs[i].oper = ToOperation(outputs[i].node());
    tf_outputs[i].index = outputs[i].index();
  }
}

void TF_OutputListSetNumOutputs(TF_OutputList* o, int num_outputs,
                                TF_Status* s) {
  unwrap(o)->expected_num_outputs = num_outputs;
  unwrap(o)->outputs.clear();
  unwrap(o)->outputs.resize(num_outputs);
}
int TF_OutputListNumOutputs(TF_OutputList* o) {
  return unwrap(o)->outputs.size();
}
TF_AbstractTensor* TF_OutputListGet(TF_OutputList* o, int i) {

  tape.RecordOperation(inputs, {exp_output}, gradient_function.release());
  TF_RETURN_IF_ERROR(tape.ComputeGradient(ctx,
                                          /*targets*/ {exp_output},
                                          /*sources=*/inputs,
                                          /*output_gradients=*/{},
                                          /*result=*/outputs));
  exp_output->Unref();
  return absl::OkStatus();
}

    TF_RETURN_IF_ERROR(tape.ComputeGradient(ctx, /*targets=*/temp_outputs,
                                            /*sources=*/inputs,
                                            /*output_gradients=*/{}, outputs));
    for (auto temp_output : temp_outputs) {
      temp_output->Unref();
    }
    return absl::OkStatus();
  };
}

}  // namespace internal

  status->status = c.Run(op_reg_data->shape_inference_fn);
  if (!status->status.ok()) return;

  // Set output_shapes.
  TF_ShapeAndTypeList* output_shapes_result =
      TF_NewShapeAndTypeList(c.num_outputs());
  for (int i = 0; i < c.num_outputs(); ++i) {
    ShapeHandle shape_handle = c.output(i);
    TF_ShapeAndType& shape = output_shapes_result->items[i];
    shape.num_dims = c.Rank(shape_handle);

  auto outputs = parallel_device.Execute(
      context.get(), {reduced_values.get(), run_collective.get()},
      "AssertAndCollective", TFE_OpGetAttrs(call_op.get()),
      /*expected_max_outputs=*/1, status.get());
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());
  ASSERT_EQ(outputs->size(), 1);
  ParallelTensor* parallel_result = (*outputs)[0].get();

                absl::Span<AbstractTensorHandle*> outputs) {
  return ops::AddV2(ctx, inputs[0], inputs[1], &outputs[0], "Add");
}

Status ExpModel(AbstractContext* ctx,
                absl::Span<AbstractTensorHandle* const> inputs,
                absl::Span<AbstractTensorHandle*> outputs) {
  return ops::Exp(ctx, inputs[0], &outputs[0], "Exp");
}

Status SqrtModel(AbstractContext* ctx,

 public:
  Status Compute(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    for (int i = 0; i < grad_outputs.size(); i++) {
      auto grad_input = grad_outputs[i];
      // TODO(srbs): Should we add a copy contructor to AbstractTensorHandle
      // that takes care of this similar to `Tensor`?

void TestRemoteExecuteSilentCopiesOp(bool async, bool remote,
                                     bool remote_func_outputs = false) {
  return TestRemoteExecuteSilentCopies(async, remote, /*func=*/false,
                                       /*heavy_load_on_streaming_rpc=*/false,
                                       remote_func_outputs);
}

TEST(CAPI, RemoteExecuteSilentCopies) {