BLUR

该数据集为大型语言模型的遗忘学习提供了一个基准，它提供了更为真实的遗忘与保留重叠的场景，在现有遗忘学习基准的基础上有了显著的扩展，包括扩展的评价任务、结合遗忘与保留的查询，以及难度不一的重新学习数据集。此外，该基准还包含了多种遗忘学习方法以及性能评价指标，如遗忘质量、保留质量和困惑度等。这项任务旨在对大型语言模型进行全面的遗忘学习评估。

This dataset serves as a benchmark for forgetting learning in large language models (LLMs). It provides more realistic scenarios where forgetting and knowledge retention co-occur, and constitutes a substantial expansion over existing forgetting learning benchmarks. The expanded scope includes extended evaluation tasks, queries that integrate forgetting and knowledge retention, and re-learning datasets with varying levels of difficulty. Additionally, this benchmark incorporates multiple forgetting learning methods and performance evaluation metrics, such as forgetting quality, retention quality, and perplexity, among others. This benchmark is designed to facilitate comprehensive forgetting learning assessments of large language models.