selfcorrexp2/llama3_sft_balanced_rr60k_train_on_corr_ep3_full_testtmp10_vllmexp

该数据集包含多个字段，如索引（idx）、提示（prompt）、答案序列（answers）、 ground truth（gt）、代理标签（proxy_label）和奖励（rewards）等。数据集被划分为训练集，大小为42161868字节，共有15000个示例。提供了默认配置，其中包括训练数据文件的路径。具体的数据集内容和用途未在README中描述。

The dataset includes multiple fields such as index (idx), prompt, answer sequence (answers), ground truth (gt), proxy label (proxy_label), and rewards. The dataset is split into a training set, which is 42161868 bytes in size and contains 15000 examples. A default configuration is provided, which includes the path to the training data files. The specific content and purpose of the dataset are not described in the README.