*   To explicitly enable or disable oneDNN optimizations, set the environment variable `TF_ENABLE_ONEDNN_OPTS` to `1` (enable) or `0` (disable) before running TensorFlow. To fall back to default settings, unset the environment variable.
    *   oneDNN optimizations can yield slightly different numerical results compared to when oneDNN optimizations are disabled due to floating-point round-off errors from

  auto pop_call_state = gtl::MakeCleanup([this] { this->call_state_.pop(); });

  std::vector<Gradient*> forward_grads;
  if (forward_function == nullptr) {
    // We have no special-cased forward gradient. Fall back to running the
    // backward function under a gradient tape.
    forward_grads.resize(output_tensors.size());
    TF_RETURN_IF_ERROR(ForwardpropFromTape(
        op_type, output_tensors, backward_function_getter,