gupta-tanish/Ultrafeedback-REFA-seed-57

该数据集是一个文本数据集，包含提示文本(prompt)、多个参与者的文本内容(A0, A1, A2, A3)及其角色信息，还有对话消息(messages)和相应的角色信息。每个参与者内容包含文本和角色两个字段。此外，每个参与者都有一个评分字段。数据集分为训练和测试两部分，每个部分有多个子集，包括偏好数据(pref)、软分数训练数据(sft)和生成数据(gen)。

The dataset is a text dataset containing prompt text, text content and role information from multiple participants (A0, A1, A2, A3), messages with corresponding role information. Each participants content includes two fields: text and role. In addition, there is a score field for each participant. The dataset is divided into training and testing parts, each with several subsets, including preference data (pref), soft score training data (sft), and generated data (gen).