OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Models

TL;DR

OBLIVIATE introduces a robust, efficient framework for unlearning specific data from large language models, effectively removing targeted information while maintaining overall model utility and resisting attacks.

Contribution

The paper presents a novel unlearning method using structured fine-tuning with LoRA, improving efficiency and robustness in removing sensitive data from LLMs.

Findings

Effective removal of targeted data demonstrated across multiple datasets.

Maintains high model utility and fluency after unlearning.

Resists membership inference attacks successfully.

Abstract

Large language models (LLMs) trained over extensive corpora risk memorizing sensitive, copyrighted, or toxic content. To address this, we propose \textbf{OBLIVIATE}, a robust unlearning framework that removes targeted data while preserving model utility. The framework follows a structured process: extracting target tokens, building retain sets, and fine-tuning with a tailored loss function comprising three components -- masking, distillation, and world fact. Using low-rank adapters (LoRA) ensures efficiency without compromising unlearning quality. We conduct experiments on multiple datasets, including Harry Potter series, WMDP, and TOFU, using a comprehensive suite of metrics: \emph{forget quality} (via a new document-level memorization score), \emph{model utility}, and \emph{fluency}. Results demonstrate its effectiveness in resisting membership inference attacks, minimizing the impact on retained data, and maintaining robustness across diverse scenarios.