trying to understand how llm , fine-tuning works, etc