KnowUnDo

该数据集是一个包含版权内容和用户隐私领域的大型基准测试集，旨在评估大型语言模型中的知识遗忘效果。该数据集用于衡量知识遗忘方法的有效性，并包含与版权和用户隐私相关的敏感数据，以便对知识遗忘评估任务进行深入研究和实践。

This dataset is a large-scale benchmark set encompassing copyrighted content and user privacy-related fields, which aims to evaluate the knowledge forgetting effect in large language models (LLMs). It is used to quantify the effectiveness of knowledge forgetting methods, and contains sensitive data related to copyright and user privacy to facilitate in-depth research and practical applications of the knowledge forgetting evaluation task.

搜集汇总

背景与挑战

背景概述

KnowUnDo是一个用于大语言模型知识遗忘评估的数据集，专注于版权和用户隐私领域，包含copyright和privacy两种配置，每种配置提供unlearn（遗忘）和retention（保留）分割，旨在帮助研究模型在遗忘敏感信息时是否影响必要知识。