yufan/Preference_Dataset_Merged
收藏Hugging Face2024-12-17 更新2024-12-14 收录
下载链接：
https://hf-mirror.com/datasets/yufan/Preference_Dataset_Merged
下载链接
链接失效反馈官方服务：
资源简介：
该数据集由多个开源偏好数据集组成，包括Arena Human Preference、Anthropic HH、MT-Bench Human Judgement、Ultra Feedback和Tulu3 Preference Dataset。数据集的特征包括prompt、chosen、rejected和source，其中chosen和rejected是包含content和role的列表。数据集经过多种清理方法处理，包括语言检测、去重、长度限制和移除安全相关部分。数据集已用于训练Mistral-Nemo-Base-2407奖励模型，并用于清理SFT数据集。
This dataset collects famous preference datasets and converts them into a unified format. The sources of the dataset include multiple datasets on Hugging Face. The features of the dataset include prompt, chosen, rejected, and source, where chosen and rejected are lists containing content and role. The dataset is divided into a training set, containing 540381 samples.
提供机构：
yufan
