It's a little too late for that, all the models train on prompts and responses.