Sequential Model Editing on ZsRE T = 5000
98.9EfficacyBetaEdit
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BetaEditBackbone=GPT-J-6B2026.05 | 98.9 | 95.1 | 26.7 | |
| BetaEditBackbone=LLaMA3-8B-Instruct2026.05 | 97.4 | 93.2 | 44.2 | |
| MEMITBackbone=LLaMA3-8B-Instruct2026.05 | 97 | 93.1 | 43.5 | |
| BetaEditBackbone=Qwen3-4B-Instruct2026.05 | 96.7 | 90.9 | 38.1 | |
| AlphaEditBackbone=GPT-J-6B2026.05 | 96.5 | 92 | 26 | |
| MEMITBackbone=Qwen3-4B-Instruct2026.05 | 95.5 | 89.9 | 38.9 | |
| MEMITBackbone=GPT-J-6B2026.05 | 95 | 89 | 24 | |
| EMMETBackbone=GPT-J-6B2026.05 | 94.3 | 87.5 | 26.7 | |
| PMETBackbone=LLaMA3-8B-Instruct2026.05 | 91 | 87 | 41 | |
| AlphaEditBackbone=Qwen3-4B-Instruct2026.05 | 89.1 | 84 | 37.4 | |
| PMETBackbone=GPT-J-6B2026.05 | 89 | 81 | 28 | |
| PMETBackbone=Qwen3-4B-Instruct2026.05 | 88.2 | 80.3 | 35.5 | |
| AlphaEditBackbone=LLaMA3-8B-Instruct2026.05 | 84.2 | 79 | 39.3 | |
| Pre-editedBackbone=LLaMA3-8B-Instruct2026.05 | 37.2 | 36.6 | 38.5 | |
| Pre-editedBackbone=Qwen3-4B-Instruct2026.05 | 32.3 | 32.6 | 35.1 | |
| Pre-editedBackbone=GPT-J-6B2026.05 | 27 | 26.2 | 27 |