tensors.reserve(output_tensors.size());
  for (const TapeTensor& o : output_tensors) {
    // Note: the tensor can have already been watched and hence be in the tape,
    // so we cannot check that we're inserting it here.
    tensor_tape_[o.GetID()] = op_id;
    tensor_usage_[o.GetID()] = 1;
    tensors.push_back(o);
  }
  op_tape_[op_id] = OpTapeEntry<BackwardFunction, TapeTensor>{

    TFE_Executor*, TF_Status* status);

// When an error happens, any pending operations are discarded, and newly issued
// ops return an error. This call clears the error state and re-enables
// execution of newly issued ops.
//
// Note that outputs of discarded ops remain in a corrupt state and should not
// be used for future calls.

    tensorflow::ImmediateExecutionTensorHandle* unwrapped_handle =
        tensorflow::unwrap(handles[i]);
    if (tensorflow::CustomDeviceTensorHandle::classof(unwrapped_handle)) {
      // One of the inputs we're trying to pack is on a custom device. We'll let
      // the first custom device we see handle all of the packing.
      auto* custom_device_handle =
          tensorflow::down_cast<tensorflow::CustomDeviceTensorHandle*>(

  // allocation/scheduling.
  //
  // TODO(allenl): Keep a map from outer thread to list of inner threads rather
  // than a single list of threads so aliased nested parallel devices don't
  // re-use a thread.
  std::vector<std::unique_ptr<DeviceThread>> device_threads_;
  // A cancellation manager to use if the caller does not provide one. When ops

    *   Added a new `rerandomize_each_iteration` argument for the `tf.data.Dataset.random()` operation, which controls whether the sequence of generated random numbers should be re-randomized every epoch or not (the default behavior). If `seed` is set and `rerandomize_each_iteration=True`, the `random()` operation will produce a different (deterministic) sequence of numbers every epoch.

  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
        // each thread would cause collectives to deadlock. For consistency we
        // create a new sync executor for every thread.
        //