KETIBench

该数据集名为KETIBench，涵盖了五种有害编辑类型，这些类型包括大多数流行的有毒编辑，同时还包含了一种良性的事实性编辑。KETIBench旨在帮助识别LLMs中的不同编辑类型，并包含了多种知识编辑方法。其任务是进行知识编辑类型识别，简称Keti。

KETIBench is a dedicated dataset that encompasses five categories of harmful editing types. These categories cover most prevalent toxic editing variants, while also incorporating one type of benign factual editing. Designed to facilitate the identification of distinct editing types within Large Language Models (LLMs), KETIBench integrates multiple knowledge editing approaches. Its core task is knowledge editing type recognition, abbreviated as Keti.