Fine tune LLMs with a meta learner to escape local minima - View it on GitHub