Harry Potter Books and News Articles

该数据集包含了《哈利·波特》系列文本和各类新闻文章，旨在评估语言模型中取消学习算法的性能。此外，该数据集还用于对八种流行的取消学习算法进行基准测试，测试标准包括逐字记忆、知识记忆、隐私泄露、效用保持、可扩展性和可持续性等方面。这项任务旨在对语言模型上的机器取消学习算法进行评估。

This dataset contains texts from the Harry Potter series and a wide range of news articles. It is designed to evaluate the performance of unlearning algorithms in language models. Furthermore, it is utilized to benchmark eight prevalent unlearning algorithms, with evaluation metrics covering verbatim memorization, knowledge memorization, privacy leakage, utility retention, scalability, and sustainability, among other aspects. This task aims to evaluate machine unlearning algorithms deployed on language models.