Synthetic Held-out Data

该数据集是为了支持数据推断（DI）任务而生成的人工合成数据，其目的是尽可能贴近原始数据集的分布特征。为此，该数据集采用了包括上下文学习、偏好优化和后缀补全在内的多种方法来生成，以最小化与原始数据集分布上的偏差。其规模根据原始数据集的大小而有所不同，所针对的任务是数据推断（DI）。

This dataset is synthetic data generated to support the Data Inference (DI) task, aiming to closely match the distributional characteristics of the original dataset. To achieve this goal, multiple methods including in-context learning, preference optimization, and suffix completion are employed for its generation, so as to minimize the distributional deviation from the original dataset. The scale of the dataset varies according to the size of the original dataset, and it is specifically designed for the Data Inference (DI) task.