ai2-adapt-dev/ultrafeedback-pairs

该数据集包含多个配置，每个配置都有一个训练集，特征包括提示、选择的回答、拒绝的回答、选择的模型、拒绝的模型、选择的评分和拒绝的评分。数据集的大小和下载大小也详细列出。

The dataset contains multiple configurations, each with different features and data splits. Features include prompt, chosen content and role, rejected content and role, chosen model and rejected model, and chosen and rejected ratings. Each configuration has a training set, with data size and number of samples provided.