zhengbang0707/MCI_REFUEL_reproduce_mask_CUDA

该数据集包含了选中的文本（chosen）和拒绝的文本（reject），每个文本都有内容（content）和角色（role）两个字段。此外，还包括了文本的序列表示（token）、掩码（mask）和用户相关的掩码（mask_user），以及选中或拒绝文本的奖励（reward）和对数概率（logprob）。数据集分为训练集和测试集，训练集大小为46006905字节，包含234个样本，测试集大小为20666133字节，包含100个样本。

The dataset includes chosen and reject texts, each with content and role fields. It also contains the texts sequence representation (token), masks (mask), user-related masks (mask_user), and the reward (reward) and log probability (logprob) for chosen or rejected texts. The dataset is split into a training set and a test set, with the training set being 46006905 bytes in size and containing 234 samples, and the test set being 20666133 bytes in size and containing 100 samples.