// Method to copy a tensor from the custom device to a target device.
  TFE_TensorHandle* (*copy_tensor_from_device)(TFE_Context* context,
                                               TFE_TensorHandle* tensor,
                                               const char* target_device_name,
                                               TF_Status* status,

  // before `StartExecute` is called again. Using `StartExecute` with `Join`
  // allows the caller to schedule computation on multiple ParallelDevices
  // without sequencing those operations (first call `StartExecute` on each
  // parallel device, then call `Join` on each; even if some of the `Join`s
  // return a bad status the caller must run all of the `Join`s or any future

// The returned string is heap-allocated, and caller should call free() on it.
TF_CAPI_EXPORT extern const char* TF_GraphDebugString(TF_Graph* graph,
                                                      size_t* len);

// Returns the function content in a human-readable format, with length set in
// `len`. The format is subject to change in the future.
// The returned string is heap-allocated, and caller should call free() on it.
//

  in_grads.reserve(input_tensors.size());
  for (int target_index = 0; target_index < input_tensors.size();
       ++target_index) {
    const auto current_grad =
        accumulated_gradients_.find(input_tensors[target_index].GetID());
    if (current_grad == accumulated_gradients_.end()) {
      if (IsDtypeTrainable(input_tensors[target_index].GetDType())) {

// things like the available devices, resource manager etc.
// TFE_Context must outlive all tensor handles created using it. In other
// words, TFE_DeleteContext() must be called after all tensor handles have
// been deleted (with TFE_DeleteTensorHandle).
//
// TODO(ashankar): Merge with TF_Session?
typedef struct TFE_Context TFE_Context;

TF_CAPI_EXPORT extern TFE_Context* TFE_NewContext(

  typedef void (*MemoryReleaser)(void* data, size_t len, void* arg);

  // Create a tensor instance from the given data buffer and description.
  // `memory_releaser` will be called on destruction, and it's responsible for
  // cleaning up the underlying buffer.
  virtual AbstractTensorInterface* CreateTensor(
      DataType dtype, const int64_t* dims, int num_dims, void* data, size_t len,

  ///    placed in `out`.
  /// 4) `TF_OK` otherwise (i.e. the read succeeded, and at least one byte was
  /// placed
  ///    in `buffer`).
  ///
  /// Caller is responsible for allocating memory for `buffer`.
  /// `buffer` will be left unchanged in case of errors.
  int64_t Read(const std::string& filename, size_t offset, size_t n,
               char* buffer, TF_Status* status);