Lune

ICLR2025Top-tier venue

Towards Effective Evaluations and Comparisons for LLM Unlearning Methods

Qizhou Wang, Bo Han, Puning Yang, Jianing Zhu, Tongliang Liu, Masashi Sugiyama

Abstract

The imperative to eliminate undesirable data memorization underscores the significance of machine unlearning for large language models (LLMs). Recent research has introduced a series of promising unlearning methods, notably boosting the practical significance of the field. Nevertheless, adopting a proper evaluation framework to reflect the true unlearning efficacy is also essential yet has not received adequate attention. This paper seeks to refine the evaluation of LLM unlearning by addressing two key challenges-a) the robustness of evaluation metrics and b) the trade-offs between competing goals. The first challenge stems from findings that current metrics are susceptible to various red teaming scenarios. It indicates that they may not reflect the true extent of knowledge retained by LLMs but rather tend to mirror superficial model behaviors, thus prone to attacks. We address this issue by devising and assessing a series of candidate metrics, selecting the most robust ones under various types of attacks. The second challenge arises from the conflicting goals of eliminating unwanted knowledge while retaining those of others. This trade-off between unlearning and retention often fails to conform the Pareto frontier, rendering it subtle to compare the efficacy between methods that excel only in either unlearning or retention. We handle this issue by proposing a calibration method that can restore the original performance on non-targeted data after unlearning, thereby allowing us to focus exclusively on assessing the strength of unlearning. Our evaluation framework notably enhances the effectiveness when assessing and comparing various LLM unlearning methods, further allowing us to benchmark existing works, identify their proper hyper-parameters, and explore new tricks to enhance their practical efficacy. The code is publicly available at: https://github.com/tmlr-group/Unlearning-with-Control.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

10 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Llemma: An Open Language Model for Mathematics cites Language Models are Few-Shot Learners
  • Direct Preference Optimization: Your Language Model is Secretly a Reward Model cites Language Models are Few-Shot Learners
  • Explainable LLM Unlearning through Reasoning cites Direct Preference Optimization: Your Language Model is Secretly a Reward Model
  • Explainable LLM Unlearning through Reasoning cites The WMDP Benchmark: Measuring and Reducing Malicious Use with Unlearning
  • Explainable LLM Unlearning through Reasoning cites LLM Unlearning with LLM Beliefs
  • LLM Unlearning with LLM Beliefs cites Explainable LLM Unlearning through Reasoning
  • The WMDP Benchmark: Measuring and Reducing Malicious Use with Unlearning cites Direct Preference Optimization: Your Language Model is Secretly a Reward Model
  • LLM Unlearning with LLM Beliefs cites Direct Preference Optimization: Your Language Model is Secretly a Reward Model
  • LLM Unlearning with LLM Beliefs cites The WMDP Benchmark: Measuring and Reducing Malicious Use with Unlearning
  • LLM Unlearning with LLM Beliefs cites Do LLMs Really Forget? Evaluating Unlearning with Knowledge Correlation and Confidence Awareness
  • Constrained Entropic Unlearning: A Primal-Dual Framework for Large Language Models cites The WMDP Benchmark: Measuring and Reducing Malicious Use with Unlearning
  • Constrained Entropic Unlearning: A Primal-Dual Framework for Large Language Models cites Extracting Training Data from Large Language Models
  • Do LLMs Really Forget? Evaluating Unlearning with Knowledge Correlation and Confidence Awareness cites The WMDP Benchmark: Measuring and Reducing Malicious Use with Unlearning
  • Extracting Training Data from Large Language Models cites Language Models are Few-Shot Learners
  • WaterDrum: Watermark-based Data-centric Unlearning Metric cites The WMDP Benchmark: Measuring and Reducing Malicious Use with Unlearning
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers10

  1. LLM Unlearning with LLM BeliefsKemou Li, Qizhou Wang, Yue Wang, Fengpeng Li et al.ICLR 2026 · 20 citations
  2. Do LLMs Really Forget? Evaluating Unlearning with Knowledge Correlation and Confidence AwarenessRongzhe Wei, Peizhi Niu, Hans Hao-Hsun Hsu, Ruihan Wu et al.NeurIPS 2025 · 20 citations
  3. Constrained Entropic Unlearning: A Primal-Dual Framework for Large Language ModelsTaha Entesari, Arman Hatami, Rinat Khaziev, Anil Ramakrishna et al.NeurIPS 2025 · 12 citations
  4. Explainable LLM Unlearning through ReasoningJunfeng Liao, Qizhou Wang, Shanshan Ye, Xin Yu et al.ICLR 2026 · 8 citations
  5. WaterDrum: Watermark-based Data-centric Unlearning MetricXinyang Lu, Xinyuan Niu, Gregory Kang Ruey Lau, Nhung Bui et al.ICLR 2026 · 8 citations

Builds on22

  1. Language Models are Few-Shot LearnersTom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah et al.NeurIPS 2020 · 64,255 citations
  2. Direct Preference Optimization: Your Language Model is Secretly a Reward ModelRafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning et al.NeurIPS 2023 · 10,924 citations
  3. Extracting Training Data from Large Language ModelsNicholas Carlini, Florian Tramèr, Eric Wallace, Matthew Jagielski et al.USENIX Security 2021 · 2,866 citations
  4. Llemma: An Open Language Model for MathematicsZhangir Azerbayev, Hailey Schoelkopf, Keiran Paster, Marco Dos Santos et al.ICLR 2024 · 433 citations
  5. The WMDP Benchmark: Measuring and Reducing Malicious Use with UnlearningNathaniel Li, Alexander Pan, Anjali Gopal, Summer Yue et al.ICML 2024 · 390 citations

Related papers

  1. Catastrophic Failure of LLM Unlearning via QuantizationZhiwei Zhang, Fali Wang, Xiaomin Li, Zongyu Wu et al.ICLR 2025
  2. MUSE: Machine Unlearning Six-Way Evaluation for Language ModelsWeijia Shi, Jaechan Lee, Yangsibo Huang, Sadhika Malladi et al.ICLR 2025
  3. Beyond Single-Value Metrics: Evaluating and Enhancing LLM Unlearning with Cognitive DiagnosisYicheng Lang, Kehan Guo, Yue Huang, Yujun Zhou et al.ACL Findings 2025
  4. Forget to Know, Remember to Use: Context-Aware Unlearning for Large Language ModelsYuefeng Peng, Parnian Afshar, Megan Ganji, Thomas Butler et al.ICML 2026 · 1 citation
  5. A Closer Look at Machine Unlearning for Large Language ModelsXiaojian Yuan, Tianyu Pang, Chao Du, Kejiang Chen et al.ICLR 2025
Dusk over the sea between two cliffs drawn in fine vertical lines