sdzxc321/PKU-SafeRLHF
收藏资源简介:
PKU-SafeRLHF数据集是一个高质量的数据集,包含83.4K个偏好条目,标注了无害性和有用性两个维度。每个条目包含对一个问题的两个回答,以及基于无害性和有用性对这两个回答的偏好。数据集中的回答来自Alpaca-7B、Alpaca2-7B和Alpaca3-8B模型。每个Q-A对都被标注了一个或多个危害类别(共19个)和严重程度级别(轻微、中等、严重)。数据集的目的是帮助研究如何使模型更加安全和无害。
PKU-SafeRLHF is a high-quality dataset consisting of 83.4K preference entries annotated across two dimensions: harmlessness and helpfulness. Each entry includes two responses to a question, along with safety meta-labels and preferences for both responses based on their helpfulness and harmlessness. The responses in this dataset come from Alpaca-7B, Alpaca2-7B, and Alpaca3-8B models. Each Q-A pair is labeled with one or more harm categories (19 in total) and a severity level (minor, moderate, severe). The dataset aims to facilitate research on making models safer and less harmful.




