gupta-tanish/Ultrafeedback-REFA-seed-191

该数据集包含多个对话角色和内容的字段，以及与这些对话相关的得分。数据集分为训练和测试两部分，每一部分都包括偏好(train_prefs/test_prefs)、软训练(train_sft/test_sft)和生成(train_gen/test_gen)数据。

The dataset includes multiple fields for dialogue roles and content, as well as scores related to these dialogues. The dataset is divided into training and testing parts, each including preference (train_prefs/test_prefs), soft training (train_sft/test_sft), and generation (train_gen/test_gen) data.