Research map / Continual learning / Continual fine-tuning of LLMs
Mapping Post-Training Forgetting in Language Models at Scale
Jackson Harmon, Andreas Hochlehnert, Matthias Bethge, Ameya Prabhu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| scale model size in instruction tuning (Qwen2.5 3B to 32B) (ablation) | instruction tuning · knowledge benchmarks | chance-adjusted forgetting: 8.5 -> 6.7 | Improved |
| scale model size in domain-continual pretraining (Coder 3B to 32B) (ablation) | domain-continual pretraining · knowledge benchmarks | chance-adjusted forgetting: 6.8 -> 6.4 (peak 8.0 at 14B) | Mixed |
| use math-specialized domain-continual pretraining | Qwen2.5 · knowledge benchmarks | chance-adjusted forgetting: 12.9 (Math 7B) vs 7.6 (Coder 7B) | Hurt |
| reasoning training (SFT/RL) from base vs instruction tuning | base models · knowledge benchmarks | backward transfer | Improved |
| low-data reasoning SFT from instruct models | Qwen instruct models (s1.1, LIMO) · knowledge benchmarks | forgetting and backward transfer | No clear effect |
| merge post-trained model with base checkpoint (linear/SLERP) (ablation) | Qwen2.5-Coder-7B, OpenThinker3-7B, OpenThinker-7B · knowledge benchmarks | forgetting | Mixed |
| measure forgetting sample-wise instead of by aggregate accuracy difference (ablation) | Qwen2.5-Coder-7B · domain-continual pretraining | measured forgetting | Improved |
| chat-template prompting of base-initialized coder models instead of few-shot (ablation) | domain-continual pretrained models · knowledge benchmarks | measured forgetting | Hurt |
Closest papers by meaning
- Spurious Forgetting in Continual Learning of Language Models (ICLR 2025 · similarity 0.91)
- Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-Training (NeurIPS 2026 · similarity 0.91)
- Demystifying Language Model Forgetting with Low-rank Example Associations (NeurIPS 2025 · similarity 0.90)
- What Will My Model Forget? Forecasting Forgotten Examples in Language Model Refinement (ICML 2024 · similarity 0.90)
- How Do Large Language Models Acquire Factual Knowledge During Pretraining? (NeurIPS 2024 · similarity 0.89)
- On the Loss of Context Awareness in General Instruction Fine-tuning (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
