// Reads the value of specific cell of a counter that was created with 1 label.
TF_CAPI_EXPORT extern int64_t TFE_MonitoringReadCounter1(
    TFE_MonitoringCounterReader*, const char* label_value);

#ifdef __cplusplus
} /* end extern "C" */
#endif

                 absl::Span<const int64_t> shape, const TF_DataType dtype)
      : device_(device),
        tensors_(std::move(tensors)),
        shape_(std::vector<int64_t>(shape.begin(), shape.end())),
        dtype_(dtype) {}
  ParallelTensor(const ParallelDevice& device,
                 std::vector<TensorHandlePtr> tensors, const TF_DataType dtype)
      : device_(device),
        tensors_(std::move(tensors)),

                                                TF_Status* status);

// ----------------------------------------------------------------

#ifdef __cplusplus
} /* end extern "C" */
#endif

// TFE_ContextEndStep while ops that rely on the step container may be running.
TF_CAPI_EXPORT extern void TFE_ContextEndStep(TFE_Context* ctx);

#ifdef __cplusplus
} /* end extern "C" */
#endif

#ifdef __cplusplus
// A workaround to ease conversion to and from numpy objects and
// TFE_TensorHandle's.
//
// TODO(ashankar): Figure out an alternative scheme that precludes the need for

  for (int i = 0, end = target_tensor_ids.size(); i < end; ++i) {
    const int64_t id = target_tensor_ids[i];
    if (output_gradients.empty() || output_gradients[i] == nullptr) {
      auto tensor_it = tensor_tape.find(id);
      if (tensor_it != tensor_tape.end() && tensor_it->second != -1) {
        auto op_it = op_tape.find(tensor_it->second);
        if (op_it == op_tape.end()) {
          return errors::Internal(

TF_CAPI_EXPORT extern void TF_GraphRemoveFunction(TF_Graph* g,
                                                  const char* func_name,
                                                  TF_Status* status);

#ifdef __cplusplus
} /* end extern "C" */
#endif

 *
 * `use_function` indicates whether to use graph mode(true) or eager(false).
 *
 * `numerical_grad` is the pointer to the AbstractTensorHandle* which will
 * hold the numerical gradient data at the end of the function.
 */
Status CalcNumericalGrad(AbstractContext* ctx, Model forward,
                         absl::Span<AbstractTensorHandle* const> inputs,
                         int input_index, bool use_function,

// Return the number of outputs in the list.
int TF_OutputListNumOutputs(TF_OutputList* o);
// Return the `i`th output in the list.
TF_AbstractTensor* TF_OutputListGet(TF_OutputList* o, int i);
// Append a tensor at the end of the output list, growing its size by one.
void TF_OutputListPushBack(TF_OutputList* o, TF_AbstractTensor* tensor,
                           TF_Status*);

    std::list<std::string>::iterator lru_iterator;
  };

  bool LookupLocked(const std::string& key, T* value)
      ABSL_EXCLUSIVE_LOCKS_REQUIRED(mu_) {
    auto it = cache_.find(key);
    if (it == cache_.end()) {
      return false;
    }
    lru_list_.erase(it->second.lru_iterator);
    if (timer_seconds_() - it->second.timestamp > max_age_) {
      cache_.erase(it);
      return false;
    }

  TF_TriState auto_parallel;
  TF_TriState memory_optimization;
  TF_TriState scoped_allocator_optimization;
} TP_OptimizerConfigs;

#define TP_OPTIMIZER_CONFIGS_STRUCT_SIZE \
  TF_OFFSET_OF_END(TP_OptimizerConfigs, scoped_allocator_optimization)

// Struct for Optimizer. Plugin authors must provide an optimize function.
// Creation and deletion functions are optional.
typedef struct TP_Optimizer {