I've used this meta framework for LLM tuning, it really one of the best out there.