PEN

该数据集用于评估GPT-4在解决组合学习任务上的性能表现，包含了诸如复制、反向复制、位置编码（PE）和位置编码变量（PEV）等子任务，这些都是完成位置编码任务（PEN）所必需的基本元素。该数据集规模达到100万样本，用于训练语言模型在组合学习任务上的能力。

This dataset is developed to evaluate GPT-4's performance on compositional learning tasks. It covers subtasks including copy, reverse copy, Positional Encoding (PE), and Positional Encoding Variable (PEV), all of which are fundamental building blocks for the Positional Encoding Task (PEN). Comprising 1 million samples, this dataset is employed to train the capabilities of language models for compositional learning tasks.