- Implement `_save_pretrained` method to handle TensorDict state saving - Add `_from_pretrained` class method for loading SAC policy from files - Create utility function `find_and_copy_params` to handle parameter copying