// allow us to trace the data dependencies between operations and hence compute
// gradients.
//
// `ZerosLike` is not expected to be called and returns a nullptr. The creation
// of default zeros grads is handled by the `DefaultGradientFunction` registered
// for each op.
// TODO(srbs): We need to define `ZerosLike` here to keep the compiler happy.
// Figure out a way to avoid this.

  }

  // We may need to allocate zero inputs for trainable dtypes we don't have JVPs
  // for. Make sure they get cleaned up.
  std::vector<Gradient*> new_zeros;
  auto delete_new_zeros = gtl::MakeCleanup([&new_zeros, this] {
    for (Gradient* tensor : new_zeros) {
      this->vspace_.DeleteGradient(tensor);
    }
  });
  std::vector<Gradient*> in_grads;
  in_grads.reserve(input_tensors.size());