selfcorrexp2/type1_14k_type2_6k_type4_8b_type3_4k_cut_separate_pr

该数据集包含多个文本字段，包括选择的文本(chosen_txt)、被拒绝的文本(rejected_txt)、真实标签(gt)、选择的标签(chosen)、被拒绝的标签(rejected)、提示(prompt)以及一个浮点数字段margin。数据集分为训练集(train)，共有31787个样本，总大小为302MB。数据集的具体应用场景和背景未在README中说明。The dataset includes multiple text fields such as chosen text (chosen_txt), rejected text (rejected_txt), ground truth (gt), chosen label (chosen), rejected label (rejected), prompt, and a floating-point field margin. The dataset is split into a training set (train) with a total of 31,787 samples and a size of 302MB. The specific application scenario and background of the dataset are not described in the README.