Lune

EMNLP Findings2024

Revisiting Catastrophic Forgetting in Large Language Model Tuning

Hongyu Li, Liang Ding, Meng Fang, Dacheng Tao

Abstract

Catastrophic Forgetting (CF) means models forgetting previously acquired knowledge when learning new data. It compromises the effectiveness of large language models (LLMs) during fine-tuning, yet the underlying causes have not been thoroughly investigated. This paper takes the first step to reveal the direct link between the flatness of the model loss landscape and the extent of CF in the field of LLMs.

Based on this, we introduce the sharpnessaware minimization to mitigate CF by flattening the loss landscape. Experiments on three widely-used fine-tuning datasets, spanning different model scales, demonstrate the effectiveness of our method in alleviating CF. Analyses show that we nicely complement the existing anti-forgetting strategies, further enhancing the resistance of LLMs to CF.

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.

29 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

  • MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models cites Language Models are Few-Shot Learners
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers29

  1. PixelVLA: Advancing Pixel-level Understanding in Vision-Language-Action ModelWenqi Liang, Gan Sun, Yao He, Jiahua Dong et al.ICLR 2026 · 20 citations
  2. Agent Data Protocol: Unifying Datasets for Diverse, Effective Fine-tuning of LLM AgentsYueqi Song, Ketan Ramaneti, Zaid Sheikh, Ziru Chen et al.ICLR 2026 · 18 citations
  3. Deep sequence models tend to memorize geometrically; it is unclear whyShahriar Noroozizadeh, Vaishnavh Nagarajan, Elan Rosenfeld, Sanjiv KumarICML 2026 · 11 citations
  4. Attack via Overfitting: 10-shot Benign Fine-tuning to Jailbreak LLMsZhixin Xie, Xurui Song, Jun LuoNeurIPS 2025 · 11 citations
  5. Just-In-Time Reinforcement Learning: Continual Learning in LLM Agents Without Gradient Updatesyibo li, Zijie Lin, Ailin Deng, Xuan (Billy) Zhang et al.ICML 2026 · 11 citations

Builds on15

  1. Language Models are Few-Shot LearnersTom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah et al.NeurIPS 2020 · 64,255 citations
  2. TruthfulQA: Measuring How Models Mimic Human FalsehoodsStephanie Lin, Jacob Hilton, Owain EvansACL 2022 · 3,228 citations
  3. Sharpness-aware Minimization for Efficiently Improving GeneralizationPierre Foret, Ariel Kleiner, Hossein Mobahi, Behnam NeyshaburICLR 2021 · 1,861 citations
  4. MetaMath: Bootstrap Your Own Mathematical Questions for Large Language ModelsLonghui Yu, Weisen Jiang, Han Shi, Jincheng Yu et al.ICLR 2024 · 637 citations
  5. BatchEnsemble: an Alternative Approach to Efficient Ensemble and Lifelong LearningYeming Wen, Dustin Tran, Jimmy BaICLR 2020 · 569 citations

Related papers

  1. Make Continual Learning Stronger via C-FlatAng Bian, Wei Li, Hangjie Yuan, Chengrong Yu et al.NeurIPS 2024 · 48 citations
  2. Exploring Forgetting in Large Language Model Pre-TrainingChonghua Liao, Ruobing Xie, Xingwu Sun, Haowen Sun et al.ACL 2025 · 8 citations
  3. HFT: Half Fine-Tuning for Large Language ModelsTingfeng Hui, Zhenyu Zhang, Shuohuan Wang, Weiran Xu et al.ACL 2025
  4. Retaining by Doing: The Role of On-Policy Data in Mitigating ForgettingHoward Chen, Noam Razin, Karthik Narasimhan, Danqi ChenICML 2026
  5. TRAM: Bridging Trust Regions and Sharpness Aware MinimizationTom Sherborne, Naomi Saphra, Pradeep Dasigi, Hao PengICLR 2024 · 6 citations
Dusk over the sea between two cliffs drawn in fine vertical lines