SCIR-HI/Med-ChatGLM

Built on ChatGLM-6B, this project applies instruction-tuning using a Chinese medical QA dataset synthesized from knowledge graphs and GPT-3.5, with training conducted on A100 80GB GPUs using supervised fine-tuning. The medical knowledge base spans diseases, drugs, and clinical indicators with structured fields covering symptoms, complications, and treatment protocols, enabling the model to generate clinically-relevant responses that outperform the base model on domain-specific diagnostic questions.

1,036 stars. No commits in the last 6 months.