zjunlp/KnowEdit
收藏资源简介:
KnowEdit是一个用于大语言模型知识编辑的基准数据集,涵盖了六个任务:ZsRE、Wiki_recent、Wiki_counterfact、WikiBio、ConvSent和Sanitation。该数据集通过重新组织和扩展现有的数据集(如WikiBio、ZsRE、WikiData_Counterfact等)构建而成,旨在为大语言模型的知识编辑提供全面的评估。数据集的结构包括知识插入、知识修改和知识擦除等任务,每个任务都有相应的训练和测试数据。README还提供了如何使用EasyEdit工具加载和使用该数据集的示例代码,并详细说明了不同数据集的加载方式和数据结构的差异。
KnowEdit is a benchmark dataset dedicated to knowledge editing for large language models (LLMs). It covers six tasks: ZsRE, Wiki_recent, Wiki_counterfact, WikiBio, ConvSent, and Sanitation. Constructed by reorganizing and expanding existing datasets such as WikiBio, ZsRE, WikiData_Counterfact and others, this dataset aims to provide comprehensive evaluations for knowledge editing of LLMs. The dataset's structure includes tasks like knowledge insertion, knowledge modification and knowledge erasure, with corresponding training and test data for each task. The accompanying README also provides sample code for loading and using this dataset with the EasyEdit tool, and elaborates on the differences in loading methods and data structures across the various datasets.
数据集概述
本数据集用于支持论文《A Comprehensive Study of Knowledge Editing for Large Language Models》的研究。通过使用EasyEdit工具,用户可以加载并应用此基准数据集。



