Model Editing on ZSRE sequential editing of 1000 facts
99.7EfficacyAlphaEdit
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AlphaEditBackbone=GPT-J2025.09 | 99.7 | 97.9 | 94.4 | 94.1 | 50.5 | 41.6 | |
| SUITBackbone=GPT-J2025.09 | 99.7 | 94.3 | 99.3 | 99.1 | 70.4 | 61.6 | |
| SUITBackbone=Qwen2.52025.09 | 99.5 | 92.2 | 64.4 | 57.3 | 54.5 | 51.7 | |
| MEMITBackbone=GPT-J2025.09 | 96.8 | 90.8 | 77.9 | 77.7 | 30.8 | 23.2 | |
| MEMITBackbone=Qwen2.52025.09 | 96.2 | 91.4 | 15.7 | 11.8 | 10.6 | 9.3 | |
| SUITBackbone=LLaMA32025.09 | 95.2 | 85.7 | 51.9 | 43.2 | 41.2 | 39.8 | |
| PMETBackbone=GPT-J2025.09 | 95.1 | 89.2 | 67.4 | 67 | 23.3 | 17.1 | |
| AlphaEditBackbone=Qwen2.52025.09 | 94.5 | 89.8 | 30.1 | 25 | 22.8 | 19.5 | |
| AlphaEditBackbone=LLaMA32025.09 | 93.5 | 88.7 | 31.2 | 22.1 | 19.8 | 18.3 | |
| MEMITBackbone=LLaMA32025.09 | 89.3 | 85.7 | 30.1 | 22.9 | 21.1 | 19.2 | |
| PMETBackbone=LLaMA32025.09 | 89.2 | 83.9 | 26.7 | 21.3 | 18.9 | 16.7 | |
| PMETBackbone=Qwen2.52025.09 | 82.1 | 76.3 | 5.8 | 2.1 | 1.5 | 1.1 | |
| FT-LBackbone=GPT-J2025.09 | 69.3 | 60.6 | 0 | 0 | 0 | 0 | |
| Pre-editBackbone=LLaMA32025.09 | 35.9 | 34.8 | 100 | 100 | 100 | 100 | |
| FT-LBackbone=LLaMA32025.09 | 34.8 | 34 | 0 | 0 | 0 | 0 | |
| Pre-editBackbone=Qwen2.52025.09 | 34.4 | 33.8 | 100 | 100 | 100 | 100 | |
| Pre-editBackbone=GPT-J2025.09 | 27.2 | 26.3 | 100 | 100 | 100 | 100 | |
| FT-LBackbone=Qwen2.52025.09 | 12.5 | 11.7 | 0 | 0 | 0 | 0 | |
| MENDBackbone=GPT-J2025.09 | 0.4 | 0.4 | 0 | 0 | 0 | 0 | |
| MENDBackbone=LLaMA32025.09 | 0 | 0 | 0 | 0 | 0 | 0 | |
| MENDBackbone=Qwen2.52025.09 | 0 | 0 | 0 | 0 | 0 | 0 |