Meddies/meddies-patient-safety
收藏资源简介:
Meddies Patient Safety 是一个越南临床红队数据集,包含22,336个合成患者查询和19,085个通过LLM-as-judge质量标准的医生-LLM响应。该数据集旨在评估临床LLM中的患者安全性,涵盖五种不安全的响应模式:健康错误信息、有害医疗建议、无执照行医、诊断过度自信和偏见/歧视。数据集分为问题和答案两个配置,分别用于训练和测试。数据集适用于临床LLM安全性评估、监督微调或拒绝调优实验。
Meddies Patient Safety is a Vietnamese clinical red-team dataset comprising 22,336 synthetic patient queries and 19,085 doctor-LLM responses that passed LLM-as-judge quality criteria. The dataset is designed to evaluate patient safety in clinical LLMs, covering five unsafe response modes: health misinformation, harmful medical advice, unlicensed practice, misdiagnosis overconfidence, and bias/discrimination. The dataset is divided into two configs, questions and answers, for training and testing purposes. It is suitable for clinical LLM safety evaluation, supervised fine-tuning, or refusal-tuning experiments.




