absl::InlinedVector<const void*, 4> values_vector;
        values_vector.reserve(s_size);
        absl::InlinedVector<size_t, 4> lengths_vector;
        lengths_vector.reserve(s_size);
        for (int i = 0; i < s_size; ++i) {
          const string& v = default_value.list().s(i);
          values_vector.push_back(v.data());
          lengths_vector.push_back(v.size());
        }

  Status FileExists(const std::string& fname, TransactionToken* token) override;
  bool FilesExist(const std::vector<std::string>& files,
                  TransactionToken* token,
                  std::vector<Status>* status) override;
  Status GetChildren(const std::string& dir, TransactionToken* token,
                     std::vector<std::string>* result) override;
  Status GetMatchingPaths(const std::string& pattern, TransactionToken* token,

      }
    }
  }

 private:
  // TODO(b/174778737): Only hold needed inputs.
  vector<AbstractTensorHandle*> forward_inputs_;
};

class DivNoNanGradientFunction : public GradientFunction {
 public:
  explicit DivNoNanGradientFunction(vector<AbstractTensorHandle*> f_inputs,
                                    vector<AbstractTensorHandle*> f_outputs)
      : forward_inputs_(f_inputs), forward_outputs_(f_outputs) {

    std::vector<TF_Output> out;
    for (auto op : ops) {
      out.push_back({op, 0});
    }
    return out;
  }

  void Define(int num_opers, const std::vector<TF_Operation*>& opers,
              const std::vector<TF_Operation*>& inputs,
              const std::vector<TF_Operation*>& outputs,
              const std::vector<string>& output_names,
              bool expect_failure = false) {

  virtual bool UsesTFRT() = 0;

  // List attributes of available devices
  virtual void ListDevices(std::vector<DeviceAttributes>* devices) = 0;

  // Add `devices` into context's device manager. Context's device manager
  // will take ownership and maintain devices' lifetime.
  virtual Status AddDevices(std::vector<std::unique_ptr<Device>> devices) = 0;

  // Block until all pending nodes are finished.

limitations under the License.
==============================================================================*/

#include <memory>
#include <utility>
#include <vector>

#include "absl/strings/str_cat.h"
#include "tensorflow/c/c_api.h"
#include "tensorflow/c/eager/abstract_context.h"
#include "tensorflow/c/eager/c_api_internal.h"

    absl::flat_hash_set<int> null_indices;
    {
      AbstractContextPtr func_ctx(BuildFunction(fn_name));
      std::vector<AbstractTensorHandle*> func_inputs;
      func_inputs.reserve(inputs.size());
      TF_RETURN_IF_ERROR(
          CreateParamsForInputs(func_ctx.get(), inputs, &func_inputs));
      std::vector<AbstractTensorHandle*> model_outputs;
      model_outputs.resize(outputs.size());

  }

  // Checks the expected result of shape inference for the given `op`.
  void CheckOutputShapes(
      TFE_Op* op,
      const std::vector<absl::optional<std::vector<int64_t>>>& input_shapes_vec,
      const std::vector<TF_Tensor*>& input_tensors,
      const absl::optional<std::vector<int64_t>>& expected_shape) {
    // Create input_shapes.
    TF_ShapeAndTypeList* input_shapes =

// gradient registerer to instantiate a GradientFunction.
struct ForwardOperation {
 public:
  string op_name;
  std::vector<AbstractTensorHandle*> inputs;
  std::vector<AbstractTensorHandle*> outputs;
  std::vector<int64_t> skip_input_indices;
  AttrBuilder attrs;
};

using GradientFunctionFactory =
    std::function<GradientFunction*(const ForwardOperation& op)>;

  EXPECT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);

  TFE_TensorHandle* h0_task0 = TestMatrixTensorHandle(ctx);
  TFE_TensorHandle* h1_task0 = TestMatrixTensorHandle(ctx);
  std::vector<TFE_TensorHandle*> handles_task0;
  if (heavy_load_on_streaming_rpc) {
    // Send 50 tensor copy requests to simulate that there have been some RPC
    // requests been enqueued.
    for (int i = 0; i < 50; ++i) {