Research map / Large language models / Retrieval and knowledge
Massive Editing for Large Language Models Based on Dynamic Weight Generation
Wentao Wan, Qiqing Lao, Zhiwei Xie, Hefeng Wu, Runnan Lin, Liang Lin, Keze Wang
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 |
|---|---|---|---|
| add InfoNCE contrastive fine-tuning of the text encoder instead of frozen BERT (ablation) | MeG · CounterFact GPT-J 1024 edits | Generality AG: 18.46 -> 84.96 | Improved |
| use InfoNCE instead of MSE fine-tuning of the text encoder (ablation) | MeG · CounterFact GPT-J 1024 edits | Generality AG: 56.54 -> 84.96 | Improved |
| add the Familiarity Network (ablation) | MeG · ZsRE Phi-2 1024 edits | Score: 81.32 -> 91.30 | Improved |
| add the Familiarity Network (ablation) | MeG · ZsRE Phi-2 1024 edits | Locality AG: 53.81 -> 88.67 | Improved |
| add the Familiarity Network (ablation) | MeG · ZsRE Phi-2 1024 edits | Generality AG: 83.50 -> 80.37 | Hurt |
| use DiT instead of MLP weight generator (ablation) | MeG · ZsRE Llama-3 10000 edits | Score: 76.21 -> 83.90 | Improved |
| use DiT instead of MLP weight generator (ablation) | MeG · ZsRE Llama-3 10000 edits | Generality AG: 45.49 -> 61.33 | Improved |
| use DiT instead of MLP weight generator (ablation) | MeG · ZsRE Llama-3 1024 edits | Locality AG: 81.64 -> 78.71 | Hurt |
| reduce diffusion steps from 1000 to 50 (ablation) | MeG · ZsRE Phi-2 10000 edits | Score: 82.80 -> 82.51 | No clear effect |
| reduce diffusion steps from 1000 to 10 (ablation) | MeG · ZsRE Phi-2 10000 edits | Score: 82.80 -> 77.83 | Hurt |
| reduce diffusion steps from 1000 to 50 (inference cost) (ablation) | MeG · ZsRE Phi-2 10000 edits | diffusion time per edit (s): 0.304 -> 0.031 | Improved |
| increase edit count from 1024 to 10000 (ablation) | MeG w/o Familiarity Network · ZsRE Phi-2 | Score: 81.32 -> 73.09 | Hurt |
Closest papers by meaning
- MELO: Enhancing Model Editing with Neuron-Indexed Dynamic LoRA (AAAI 2024 · similarity 0.90)
- Precise Localization of Memories: A Fine-grained Neuron-level Knowledge Editing Technique for LLMs (ICLR 2025 · similarity 0.89)
- WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Models (NeurIPS 2024 · similarity 0.89)
- Editing Memories Through Few Targeted Neurons (AAAI 2025 · similarity 0.89)
- Edit Less, Achieve More: Dynamic Sparse Neuron Masking for Lifelong Knowledge Editing in LLMs (NeurIPS 2025 · similarity 0.89)
- Golden Layers and Where to Find Them: Improved Knowledge Editing for Large Language Models Via Layer Gradient Analysis (NeurIPS 2026 · 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.
