Sequential Model Editing on CounterFact T = 5000
96.6EfficacyBetaEdit
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BetaEditBackbone=GPT-J-6B2026.05 | 96.6 | 52.8 | 6.8 | |
| AlphaEditBackbone=GPT-J-6B2026.05 | 95.9 | 50.9 | 6.9 | |
| BetaEditBackbone=Qwen3-4B-Instruct2026.05 | 90.3 | 60.1 | 8 | |
| BetaEditBackbone=LLaMA3-8B-Instruct2026.05 | 89 | 62.4 | 15.1 | |
| MEMITBackbone=LLaMA3-8B-Instruct2026.05 | 87.1 | 65.4 | 14.4 | |
| MEMITBackbone=Qwen3-4B-Instruct2026.05 | 85.3 | 58.3 | 8.1 | |
| PMETBackbone=GPT-J-6B2026.05 | 85 | 50 | 7.8 | |
| PMETBackbone=Qwen3-4B-Instruct2026.05 | 82 | 55.2 | 7.2 | |
| PMETBackbone=LLaMA3-8B-Instruct2026.05 | 82 | 60.3 | 12.8 | |
| MEMITBackbone=GPT-J-6B2026.05 | 74.5 | 40.4 | 7.6 | |
| Pre-editedBackbone=GPT-J-6B2026.05 | 0.4 | 0.5 | 14.3 | |
| Pre-editedBackbone=Qwen3-4B-Instruct2026.05 | 0.3 | 0.5 | 13.9 | |
| Pre-editedBackbone=LLaMA3-8B-Instruct2026.05 | 0.3 | 0.3 | 21.6 |