Syghmon/rich_trajectories

该数据集包含了数据源、提示内容（包括内容和角色）、能力、奖励模型（包括地面真实值和风格）以及额外信息（包括索引、不匹配证据、源滚动ID和分割信息）。数据集分为训练集train1和train2，以及测试集test，每个集合的大小和示例数量不同。

The dataset includes data source, prompt content (including content and role), ability, reward model (including ground truth and style), and extra information (including index, misalignment evidence, source rollout ID, and split). The dataset is divided into training sets train1 and train2, and the test set test, with different sizes and numbers of examples for each set.