stalaei/sdft-medical-distil

SDFT Medical数据集是一个专为医疗问答和文本生成任务设计的NLP数据集，基于HuatuoGPT-o1医疗QA训练集，用于自我蒸馏微调（SDFT）研究。数据集包含5000个训练样本，来源于FreedomIntelligence/medical-o1-reasoning-SFT的英文部分。每个样本包括prompt（学生接收的聊天格式提示）、teacher_prompt（教师接收的聊天格式提示，包含上游数据集的黄金自由形式回答作为上下文演示）、answer（上游数据集的简短最终医疗答案）和chat_template_kwargs（用于tokenizer.apply_chat_template的额外参数）。数据集还提供了加载、评分和本地复现的详细方法，并继承了上游数据集的许可证。

The SDFT Medical dataset is an NLP dataset designed for medical question-answering and text-generation tasks, based on the HuatuoGPT-o1 medical-QA training set and used for Self-Distillation Fine-Tuning (SDFT) research. The dataset contains 5,000 training samples sourced from the English split of FreedomIntelligence/medical-o1-reasoning-SFT. Each sample includes prompt (chat-format prompt for the student), teacher_prompt (chat-format prompt for the teacher with the golden free-form Response from the upstream dataset as an in-context demonstration), answer (the short final medical answer from the upstream dataset), and chat_template_kwargs (extra kwargs for tokenizer.apply_chat_template). The README also provides detailed methods for loading, scoring, and local reproduction, and the dataset inherits the license of the upstream datasets.