selfcorrexp2/llama31_14w_wrong_6w_corr_20k_test_old_format_no_self_corr

该数据集包含了多个特征字段，如索引(idx)、是否为第一轮(first_round)、真实标签(gt)、奖励(rewards)、我的解决方案(my_solu)、标志(flag)、轮数(turn)以及会话(conversations)。会话又包括内容(content)和角色(role)。数据集分为训练集(train)，共有20000个示例，大小为84704003.17655128字节。数据集的总下载大小为34533477字节。The dataset contains multiple feature fields such as index (idx), whether its the first round (first_round), ground truth label (gt), rewards (rewards), my solution (my_solu), flag (flag), turn (turn), and conversations. Conversations include content (content) and role (role). The dataset is split into a training set (train) with a total of 20,000 examples, occupying 84704003.17655128 bytes. The total download size of the dataset is 34533477 bytes.