allenai/WildChat-1M
收藏资源简介:
WildChat数据集包含了100万条人类用户与ChatGPT之间的对话,以及相关的用户人口统计数据,如州、国家、哈希IP地址和请求头信息。数据集中的对话涵盖了广泛的用户与ChatGPT的交互场景,包括模糊的用户请求、代码切换、话题切换、政治讨论等。数据集中的对话来自GPT-3.5和GPT-4模型,其中25.53%的对话来自GPT-4。数据集还包含了OpenAI Moderation和Detoxify的审核结果,用于标记对话中的毒性内容。此外,数据集中的个人信息已经通过Microsoft Presidio和作者手写的规则进行了去标识化处理。
WildChat-1M is a dataset containing 1 million conversations between human users and ChatGPT, accompanied by demographic data including state, country, hashed IP addresses, and request headers. The dataset is multi-lingual, covering interactions in 68 languages. It includes both toxic and non-toxic user inputs and ChatGPT responses, encompassing a wide range of user behaviors and interactions such as ambiguous requests, code-switching, topic-switching, and political discussions. The data has been de-identified to protect personal and sensitive information. The dataset is suitable for instructional fine-tuning and studying user behaviors.




