Status RunAndMaybeSum(AbstractContext* ctx, Model forward,
                      absl::Span<AbstractTensorHandle* const> inputs,
                      absl::Span<AbstractTensorHandle*> outputs,
                      bool use_function) {
  AbstractTensorHandle* model_outputs[1];

  // Run the model.
  TF_RETURN_IF_ERROR(
      RunModel(forward, ctx, inputs, model_outputs, use_function));

namespace gradients {
namespace internal {
namespace {
using std::vector;
using tensorflow::TF_StatusPtr;
using tracing::TracingOperation;

class CppGradients
    : public ::testing::TestWithParam<std::tuple<const char*, bool, bool>> {
 protected:
  void SetUp() override {
    TF_StatusPtr status(TF_NewStatus());
    TF_SetTracingImplementation(std::get<0>(GetParam()), status.get());
    Status s = StatusFromTF_Status(status.get());

// deadlocks.
class DeviceThread {
 public:
  // Starts a background thread waiting for `StartExecute`.
  explicit DeviceThread(const std::string& device, const bool is_async,
                        const int in_flight_nodes_limit)
      : status_(TF_NewStatus()),
        // If the context's default exector is set to async, re-using that in

  std::unique_ptr<TF_DeviceList, decltype(&TF_DeleteDeviceList)> devices(
      TFE_ContextListDevices(context.get(), status.get()), TF_DeleteDeviceList);
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());
  bool has_tpu = false;
  for (int device_index = 0; device_index < TF_DeviceListCount(devices.get());
       ++device_index) {
    std::string device_type =
        TF_DeviceListType(devices.get(), device_index, status.get());

}

TF_Function* TF_FunctionImportFunctionDef(const void* proto, size_t proto_len,
                                          TF_Status* status) {
  tensorflow::FunctionDef fdef;
  bool success = fdef.ParseFromArray(proto, proto_len);
  if (!success) {
    status->status = InvalidArgument(
        "Invalid FunctionDef given to TF_FunctionImportFunctionDef");
    return nullptr;
  }

// functionality of `gradient_checker` by comparing the result with expected
// manual user-provided gradients.
class GradientCheckerTest
    : public ::testing::TestWithParam<std::tuple<const char*, bool, bool>> {
 protected:
  void SetUp() override {
    TF_StatusPtr status(TF_NewStatus());
    TF_SetTracingImplementation(std::get<0>(GetParam()), status.get());

    {
      Status s = StatusFromTF_Status(status.get());

                   "    }"),
      &def));
  return def.SerializeAsString();
}

void TestRemoteExecuteSilentCopies(bool async, bool remote, bool func,
                                   bool heavy_load_on_streaming_rpc,
                                   bool remote_func_outputs,
                                   bool has_packed_input) {
  CHECK(!has_packed_input || func);
  tensorflow::ServerDef server_def = GetServerDef(3);

                         int num_values) override {
    std::unique_ptr<bool[]> b(new bool[num_values]);
    for (int i = 0; i < num_values; ++i) {
      b[i] = values[i];
    }
    op_->node_builder.Attr(attr_name,
                           ArraySlice<const bool>(b.get(), num_values));

    return absl::OkStatus();
  }

typedef struct GCSFile {
  const std::string path;
  const bool is_cache_enable;
  const uint64_t buffer_size;
  ReadFn read_fn;
  absl::Mutex buffer_mutex;
  uint64_t buffer_start ABSL_GUARDED_BY(buffer_mutex);
  bool buffer_end_is_past_eof ABSL_GUARDED_BY(buffer_mutex);
  std::string buffer ABSL_GUARDED_BY(buffer_mutex);

  GCSFile(std::string path, bool is_cache_enable, uint64_t buffer_size,
          ReadFn read_fn)

class UnifiedAPI
    : public ::testing::TestWithParam<std::tuple<const char*, bool, bool>> {
 protected:
  void SetUp() override {
    TF_StatusPtr status(TF_NewStatus());
    TF_SetTracingImplementation(std::get<0>(GetParam()), status.get());
    Status s = StatusFromTF_Status(status.get());
    CHECK_EQ(errors::OK, s.code()) << s.message();
  }

 public:
  bool UseMlir() const { return strcmp(std::get<0>(GetParam()), "mlir") == 0; }