ali5341/pubmedqa-chat-format

PubMedQA（聊天格式准备）数据集是基于qiaojin/PubMedQA原始数据集的聊天格式预处理版本，专为生物医学问答的监督微调（SFT）设计。该数据集包含训练集、验证集和统计文件，以及一个预处理脚本。原始数据集专注于生物医学研究问答，答案类型为yes/no/maybe，并基于摘要进行决策。预处理过程中采用了标签优先策略，确保监督数据的清洁性，并过滤了无效的最终决策行。数据集的格式为OpenAI风格的messages格式，每条记录包含用户指令、问题和上下文，以及助手的回答和元数据。

PubMedQA (Chat-Format Preparation) is a chat-format preparation of the original qiaojin/PubMedQA dataset for biomedical QA supervised fine-tuning (SFT). The dataset includes training and validation sets, a stats file, and a preparation script. The original dataset focuses on biomedical research QA with yes/no/maybe decisions based on abstracts. The preparation process follows a labeled-first policy for cleaner supervision and filters rows without valid final decisions. The dataset is formatted in the OpenAI-style messages format, with each record containing user instructions, questions, and context, along with assistant responses and metadata.