Compare a vanilla Large Language Model's performance against one that has been given a trained adapter.