shivank21/ultrafeedback_binarized

该数据集包含了提示文本(prompt)、提示ID(prompt_id)、选中的对话内容(chosen)、被拒绝的对话内容(rejected)、对话消息(messages)以及选中对话和拒绝对话的评分(score_chosen和score_rejected)。数据集分为训练和测试两部分，每个部分又细分为偏好训练(train_prefs)、偏好测试(test_prefs)、SFT训练(train_sft)、SFT测试(test_sft)、生成训练(train_gen)、生成测试(test_gen)和普通训练(train)。每个部分包含不同数量的示例和大小。

The dataset includes prompt text (prompt), prompt ID (prompt_id), chosen dialogue content (chosen), rejected dialogue content (rejected), dialogue messages (messages), and scores for chosen and rejected dialogues (score_chosen and score_rejected). The dataset is divided into training and testing sets, with each set further divided into preference training (train_prefs), preference testing (test_prefs), SFT training (train_sft), SFT testing (test_sft), generation training (train_gen), generation testing (test_gen), and general training (train). Each part contains a different number of examples and size.