ALTER: Asymmetric LoRA for Token-Entropy-Guided Unlearning of LLMs

TL;DR

ALTER introduces an efficient asymmetric LoRA-based framework for targeted unlearning in large language models, achieving high forget accuracy with minimal impact on overall model utility, addressing knowledge entanglement and computational challenges.

Contribution

This work proposes a novel asymmetric LoRA architecture for token-level unlearning, enabling efficient and precise forgetting in large language models.

Findings

Achieves over 95% forget quality on benchmarks.

Preserves over 90% of model utility.

Outperforms baseline preservation rates significantly.

Abstract

Large language models (LLMs) have advanced to encompass extensive knowledge across diverse domains. Yet controlling what a LLMs should not know is important for ensuring alignment and thus safe use. However, effective unlearning in LLMs is difficult due to the fuzzy boundary between knowledge retention and forgetting. This challenge is exacerbated by entangled parameter spaces from continuous multi-domain training, often resulting in collateral damage, especially under aggressive unlearning strategies. Furthermore, the computational overhead required to optimize State-of-the-Art (SOTA) models with billions of parameters poses an additional barrier. In this work, we present ALTER, a lightweight unlearning framework for LLMs to address both the challenges of knowledge entanglement and unlearning efficiency. ALTER operates through two phases: (I) high entropy tokens are captured and learned via the shared A matrix in LoRA, followed by (II) an asymmetric LoRA architecture that achieves a specified forgetting objective by parameter isolation and unlearning tokens within the target subdomains. Serving as a new research direction for achieving unlearning via token-level isolation in the asymmetric framework. ALTER achieves SOTA performance on TOFU, WMDP, and MUSE benchmarks with over 95% forget quality and shows minimal side effects through preserving foundational tokens. By decoupling unlearning from LLMs' billion-scale parameters, this framework delivers excellent efficiency while preserving over 90% of model utility, exceeding baseline preservation rates of 47.8-83.6%.