Research map / Large language models / Instruction and fine-tuning
Fine-tuning Done Right in Model Editing
Wanli Yang, Rui Tang, Hongyu Zang, Du Su, Qi Cao, Jingang Wang, Huawei Shen, Xueqi Cheng, Fei Sun
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 |
|---|---|---|---|
| switch from depth-first to breadth-first pipeline (ablation) | FT-M · ZsRE, LLaMA3-8B | reliability: 75.30 -> 99.70 | Improved |
| switch from depth-first to breadth-first pipeline (ablation) | AdaLoRA · ZsRE, LLaMA3-8B | reliability: 3.80 -> 90.50 | Improved |
| switch from depth-first to breadth-first pipeline (ablation) | FT-L · ZsRE, LLaMA3-8B | reliability: 0.00 -> 0.00 | No clear effect |
| switch from depth-first to breadth-first pipeline (ablation) | RoseLoRA · ZsRE, LLaMA3-8B | reliability: 0.30 -> 1.00 | No clear effect |
| use mini-batch updates instead of batch size 1 in breadth-first pipeline (ablation) | FT-M · ZsRE | capability | Improved |
| use larger batch sizes in breadth-first pipeline (ablation) | FT-M and AdaLoRA · ZsRE | generalization | Hurt |
| tune MLP down-projection in later layers instead of coarser modules (ablation) | LocFT-BF · ZsRE, 1000 edits | capability | Improved |
| tune coarser modules (entire layer, full MLP) instead of MLP up/down (ablation) | fine-tuning editing · ZsRE, 1000 edits | generalization | Improved |
| add prefix-based data augmentation (5 random prefixes) (ablation) | LocFT-BF · CounterFact, LLaMA3-8B | generalization: 33.23 -> 50.93 | Improved |
| use proportional layer position instead of default position (ablation) | LocFT-BF · ZsRE, Qwen2.5-32B | capability: 60.27 -> 64.72 | Improved |
| use full-parameter fine-tuning instead of localized tuning (ablation) | LocFT-BF · ZsRE, LLaMA3-8B (LlamaFactory) | capability: 56.03 -> 40.44 | Hurt |
| apply LocFT-BF in place of RLEdit and UltraEdit when scaling to larger models | LocFT-BF · ZsRE, Qwen2.5-14B | reliability: 8.90 -> 99.00 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AlphaEdit — AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models (ICLR 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Linearization Explains Fine-Tuning in Large Language Models (NeurIPS 2025 · similarity 0.90)
- Towards Lifelong Model Editing via Simulating Ideal Editor (ICML 2025 · similarity 0.90)
- DELIFT: Data Efficient Language model Instruction Fine-Tuning (ICLR 2025 · similarity 0.90)
- When Scaling Meets LLM Finetuning: The Effect of Data, Model and Finetuning Method (ICLR 2024 · similarity 0.90)
- Data Efficient Adaptation in Large Language Models via Continuous Low-Rank Fine-Tuning (NeurIPS 2025 · similarity 0.90)
- Rethinking Residual Distribution in Locate-then-Edit Model Editing (NeurIPS 2025 · similarity 0.90)
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.
