On RL for LLM fine-tuning

Archived