Show HN: Minimal LLM Post-Training Experiments on an 8GB GPU (SFT, DPO, GRPO) github.com 21 points by popopanda 3 days ago