Meddies/meddies-qa

Meddies QA是一个越南语医疗问答数据集，数据被重新组织为五个领域组，包含聊天模板问答行和仅用户问题库配置，用于生成工作流程。该数据集基于医疗文献（如医学文章、药物参考、疾病解释、产品描述和面向患者的教育材料）生成越南语问题和答案，旨在支持越南语医疗大型语言模型的训练，包括监督微调、答案风格适应、检索答案实验以及问题生成管道。数据集提供两类配置：qa_*配置包含两轮对话（用户问题和助手答案），用于聊天模板训练；questions_*配置仅包含用户提示，适用于生成或过滤自定义答案的工作流程。五个领域组包括：临床健康、制药、营养补充剂、消费者健康与个人护理、医疗设备与装备。数据集规模较大，总行数超过1000万，但需注意其非商业许可限制，且答案为机器生成，可能存在临床或事实错误，不应用于临床决策。Vietnamese medical question-answer data reorganized into five domain groups, with chat-template QA rows and user-only question-bank configs for generation workflows. The dataset transforms literature-grounded healthcare information (medical articles, drug references, disease explainers, product descriptions, patient-facing education material) into Vietnamese questions and answers for training runs, enabling learning of domain coverage, answer structure, and medically relevant response patterns. It is designed for Vietnamese healthcare LLM work, including supervised fine-tuning, answer-style adaptation, retrieval-answer experiments, and question-only generation pipelines. The repository exposes two families of configs: qa_* configs with two-turn chat-template rows (user question and assistant answer) for chat-template training, and questions_* configs with one-turn user prompts only for teams to generate or filter their own answers. The five domain groups are clinical_health, pharmaceuticals, nutrition_supplements, consumer_health_personal_care, and medical_devices_equipment. The dataset has over 10 million rows in total, but it is under a non-commercial license, and answers are machine-generated with potential clinical or factual errors, not intended for clinical guidance.