selfcorrexp2/llama31_first_wrong_and_10kfirst_corr_regular_norr_20k

该数据集包含了对话的索引、提示、答案序列、是否为第一轮对话、真实标签、奖励序列、解决方案序列、是否为特定标志的对话、对话轮数以及对话内容与角色信息。数据集仅包含训练集部分，共20000个示例。

The dataset includes conversation index, prompt, answer sequence, whether its the first round of conversation, ground truth label, reward sequence, solution sequence, whether its a conversation with a specific flag, the turn of the conversation, and the content and role information of the conversation. The dataset contains only the training set with a total of 20,000 examples.