To Forget or Not? Towards Practical Knowledge Unlearning for Large Language Models

TL;DR

This paper introduces KnowUnDo, a benchmark for evaluating knowledge unlearning in LLMs, and proposes MemFlex, a method that improves the precision of unlearning sensitive information while retaining general knowledge.

Contribution

The paper presents a new benchmark for assessing unlearning accuracy and introduces MemFlex, a gradient-based method for precise knowledge unlearning in large language models.

Findings

MemFlex outperforms existing unlearning methods in precision.

Existing methods often erase too much knowledge, including essential information.

KnowUnDo provides a standardized way to evaluate unlearning effectiveness.

Abstract

Large Language Models (LLMs) trained on extensive corpora inevitably retain sensitive data, such as personal privacy information and copyrighted material. Recent advancements in knowledge unlearning involve updating LLM parameters to erase specific knowledge. However, current unlearning paradigms are mired in vague forgetting boundaries, often erasing knowledge indiscriminately. In this work, we introduce KnowUnDo, a benchmark containing copyrighted content and user privacy domains to evaluate if the unlearning process inadvertently erases essential knowledge. Our findings indicate that existing unlearning methods often suffer from excessive unlearning. To address this, we propose a simple yet effective method, MemFlex, which utilizes gradient information to precisely target and unlearn sensitive parameters. Experimental results show that MemFlex is superior to existing methods in both precise knowledge unlearning and general knowledge retaining of LLMs. Code and dataset are released at https://github.com/zjunlp/KnowUnDo.