finetune ChatGLM-6B on medical QA dataset by LoRA