selfcorrexp2/llama31_first_wrong_and_20kfirst_corr_regular_norr_20k

该数据集包含了会话信息，每个会话有内容(content)和角色(role)两个维度。此外，数据集还包含了索引(idx)、提示(prompt)、答案(answers)、是否为第一轮(first_round)、真实标签(gt)、奖励(rewards)、个人解决方案(my_solu)、标志(flag)和对话轮次(turn)等字段。数据集被划分为训练集(train)，共有20000个示例，大小约为338.97MB。

The dataset consists of conversation information, with each conversation having two dimensions: content and role. In addition, the dataset includes fields such as index (idx), prompt, answers, whether it is the first round (first_round), ground truth (gt), rewards, personal solutions (my_solu), flag, and turn of conversation. The dataset is split into a training set (train) with a total of 20,000 examples, approximately 338.97MB in size.