selfcorrexp/llama3_it_distill_first_wrong

该数据集包含对话场景的特征数据，具体包括提示信息(prompt)、地面真实标签(gt)、选择的答案(chosen_ans)、消息列表(messages)和对话列表(conversations)。消息列表和对话列表都包含了内容和角色两个维度。数据集被划分为训练集(train)，共有204898个样本，数据集大小为1.66GB。

This dataset contains feature data for dialogue scenarios, including prompt, ground truth label (gt), chosen answer (chosen_ans), message list (messages), and conversation list (conversations). Both the message list and the conversation list include content and role dimensions. The dataset is split into a training set (train) with a total of 204,898 samples, and the dataset size is 1.66GB.