Sequential Model Editing on ZsRE T = 300
98.8EfficacyPMET
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PMETBackbone=LLaMA3-8B-Instruct2026.05 | 98.8 | 94.5 | 39.4 | |
| PMETBackbone=Qwen3-4B-Instruct2026.05 | 98.6 | 94 | 37.3 | |
| EMMETBackbone=Qwen3-4B-Instruct2026.05 | 98.5 | 93.2 | 36.8 | |
| EMMETBackbone=LLaMA3-8B-Instruct2026.05 | 98.5 | 94 | 40.5 | |
| MEMITBackbone=LLaMA3-8B-Instruct2026.05 | 98.5 | 94 | 39.5 | |
| EMMETBackbone=GPT-J-6B2026.05 | 98 | 92 | 28.8 | |
| PMETBackbone=GPT-J-6B2026.05 | 98 | 93 | 29.4 | |
| AlphaEditBackbone=GPT-J-6B2026.05 | 98 | 94 | 29.5 | |
| BetaEditBackbone=LLaMA3-8B-Instruct2026.05 | 98 | 94.5 | 40.3 | |
| AlphaEditBackbone=Qwen3-4B-Instruct2026.05 | 97.9 | 93.3 | 37 | |
| BetaEditBackbone=Qwen3-4B-Instruct2026.05 | 97.5 | 93 | 36.7 | |
| BetaEditBackbone=GPT-J-6B2026.05 | 97.5 | 93 | 29.6 | |
| MEMITBackbone=Qwen3-4B-Instruct2026.05 | 96.5 | 93 | 37.1 | |
| MEMITBackbone=GPT-J-6B2026.05 | 96.5 | 92 | 30.1 | |
| SimIEBackbone=LLaMA3-8B-Instruct2026.05 | 96.5 | 92 | 36 | |
| SimIEBackbone=Qwen3-4B-Instruct2026.05 | 95.2 | 90.6 | 34.3 | |
| SimIEBackbone=GPT-J-6B2026.05 | 94.5 | 88 | 27.5 | |
| PRUNEBackbone=LLaMA3-8B-Instruct2026.05 | 90 | 80 | 36 | |
| PRUNEBackbone=Qwen3-4B-Instruct2026.05 | 89.3 | 76 | 32 | |
| PRUNEBackbone=GPT-J-6B2026.05 | 88 | 74 | 25.6 | |
| AdaEditBackbone=LLaMA3-8B-Instruct2026.05 | 88 | 76 | 32 | |
| AdaEditBackbone=GPT-J-6B2026.05 | 86 | 71 | 24.8 | |
| RECTBackbone=LLaMA3-8B-Instruct2026.05 | 86 | 74 | 32 | |
| AlphaEditBackbone=LLaMA3-8B-Instruct2026.05 | 86 | 81 | 40.1 | |
| AdaEditBackbone=Qwen3-4B-Instruct2026.05 | 85 | 73 | 28.4 | |
| RECTBackbone=GPT-J-6B2026.05 | 84 | 68 | 24.7 | |
| RECTBackbone=Qwen3-4B-Instruct2026.05 | 83.4 | 70.6 | 28.7 | |
| RLEditBackbone=LLaMA3-8B-Instruct2026.05 | 82 | 60 | 28 | |
| RLEditBackbone=GPT-J-6B2026.05 | 81 | 58 | 19 | |
| RLEditBackbone=Qwen3-4B-Instruct2026.05 | 78.8 | 55.7 | 24.3 | |
| FTBackbone=LLaMA3-8B-Instruct2026.05 | 42 | 8 | 12 | |
| FTBackbone=GPT-J-6B2026.05 | 40 | 8 | 12 | |
| FTBackbone=Qwen3-4B-Instruct2026.05 | 38.5 | 7.2 | 11.3 | |
| Pre-editedBackbone=LLaMA3-8B-Instruct2026.05 | 36.5 | 36.9 | 37.5 | |
| Pre-editedBackbone=Qwen3-4B-Instruct2026.05 | 31.6 | 32.3 | 34.8 | |
| Pre-editedBackbone=GPT-J-6B2026.05 | 26.5 | 26.7 | 28 |