可能是时候开始尝试对LLM进行微调了。根据这篇文章,如果有10GB的VRAM,就可以用PEFT来训练Llama3-8B的QLoRA模型。https://mlops.community/budget-instruction-fine-tuning-of-llama-3-8b-instructon-medical-data-with-hugging-face-google-colab-and-unsloth/