resource · 2025
基于雅意大模型的中医智能预问诊
中医杂志
基于雅意大模型的中医智能预问诊 is a resource in the TCM LLM catalog (Awesome-TCM-LLM), dated 2025. It is maintained alongside the GitHub list of Traditional Chinese Medicine large language models.
Liu et al. fine-tune YAYI 2-30B with DeepSpeed ZeRO-3 and couple it to a weighted symptom-graph agent. On the simulated tests, practitioner-exam accuracy is 94.09%, licensed Chinese-pharmacy exam accuracy 78.00%, symptom-term recognition 87.50%, and meridian recognition 68.80%. In the clinical test, syndrome accuracy is 88.33%, with fewer turns and less time than attending physicians.