AIPlans/FilteredPKU-SafeRLHF

该数据集是一个对话数据集，包含对话提示和两个可能的响应。每个响应都标注了其来源和是否安全，并且详细标注了可能存在的危害类别，如国家安全威胁、侮辱行为、歧视行为等。此外，每个响应还有一个危害严重程度的评分和用于标识更好或更安全响应的ID。数据集分为训练集和测试集。

This dataset is a dialogue dataset containing dialogue prompts and two possible responses. Each response is labeled with its source and whether it is safe, and is detailedly labeled with possible harm categories such as national security threats, insulting behavior, discriminatory behavior, etc. In addition, each response has a harm severity score and an ID for identifying a better or safer response. The dataset is split into training and test sets.