Knowledge Editing on COUNTERFACT RS
100EfficacyRETS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RETSBackbone=GPT2-XL (1.5B)2025.06 | 100 | 67.9 | 78.7 | 585.1 | |
| ROMEBackbone=GPT-J (6B)2025.06 | 100 | 79.3 | 52.8 | 621 | |
| MEMITBackbone=GPT-J (6B)2025.06 | 100 | 81.1 | 66.8 | 621.5 | |
| RETSBackbone=GPT-J (6B)2025.06 | 100 | 61.4 | 81.5 | 530.6 | |
| MEMITBackbone=Qwen2.5 (7B)2025.06 | 100 | 67.3 | 64.7 | 624.4 | |
| AlphaEditBackbone=Qwen2.5 (7B)2025.06 | 100 | 83.6 | 65 | 626.3 | |
| ROMEBackbone=Qwen2.5 (14B)2025.06 | 100 | 61.3 | 56.9 | 624 | |
| MEMITBackbone=Qwen2.5 (14B)2025.06 | 100 | 79.8 | 66.7 | 625.4 | |
| ROME-LTIBackbone=Qwen2.5 (14B)2025.06 | 100 | 70.5 | 78.1 | 623.4 | |
| ROME-LTIBackbone=GPT2-XL (1.5B)2025.06 | 99.9 | 75.7 | 57.2 | 622 | |
| ROMEBackbone=GPT2-XL (1.5B)2025.06 | 99.8 | 75.4 | 45.7 | 623.4 | |
| AlphaEditBackbone=GPT2-XL (1.5B)2025.06 | 99.8 | 76.2 | 41.2 | 623.4 | |
| AlphaEditBackbone=GPT-J (6B)2025.06 | 99.8 | 81.9 | 60.4 | 621.4 | |
| ROME-LTIBackbone=GPT-J (6B)2025.06 | 99.8 | 80.7 | 82.5 | 618.8 | |
| ROMEBackbone=Qwen2.5 (7B)2025.06 | 99.8 | 82.2 | 70.7 | 625.5 | |
| ROME-CPABackbone=GPT-J (6B)2025.06 | 99.6 | 82 | 83.5 | 618.6 | |
| RETSBackbone=Qwen2.5 (7B)2025.06 | 99.5 | 42.7 | 52.9 | 540.3 | |
| ROME-LTIBackbone=Qwen2.5 (7B)2025.06 | 99.5 | 83.7 | 84.6 | 624.5 | |
| ROME-CPABackbone=Qwen2.5 (14B)2025.06 | 99.5 | 83.4 | 84.5 | 620.5 | |
| ROME-CPABackbone=GPT2-XL (1.5B)2025.06 | 98.7 | 76.5 | 72.2 | 619.7 | |
| ROME-CPABackbone=Qwen2.5 (7B)2025.06 | 96.2 | 84 | 92.2 | 623.9 | |
| MEMITBackbone=GPT2-XL (1.5B)2025.06 | 94 | 76.4 | 61.5 | 627.4 | |
| RETSBackbone=Qwen2.5 (14B)2025.06 | 92 | 73.6 | 83.2 | 591.5 |