Knowledge Editing on Counterfact (Specific Metrics)
92.96AVG ScoreLightEdit
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LightEditBackbone=LLaMA-3 (8B)2026.04 | 92.96 | 99.68 | 81.41 | 97.8 | |
| LightEditBackbone=GPT-J2026.04 | 89.93 | 99.1 | 73.2 | 97.5 | |
| LTEBackbone=GPT-J2026.04 | 87.87 | 97.75 | 87.7 | 78.15 | |
| LTEBackbone=LLaMA-3 (8B)2026.04 | 87.53 | 97.51 | 85.56 | 79.52 | |
| AlphaEditBackbone=GPT-J2026.04 | 85.63 | 93.07 | 74.77 | 89.06 | |
| RECIPEBackbone=LLaMA-3 (8B)2026.04 | 76.64 | 88.9 | 79.95 | 61.08 | |
| RECIPEBackbone=GPT-J2026.04 | 74.82 | 75.65 | 68.3 | 80.5 | |
| MEMITBackbone=GPT-J2026.04 | 67.87 | 95.79 | 44.6 | 63.23 | |
| AlphaEditBackbone=LLaMA-3 (8B)2026.04 | 62.95 | 71.44 | 64.09 | 53.33 | |
| GRACEBackbone=LLaMA-3 (8B)2026.04 | 35.67 | 6.5 | 0.5 | 100 | |
| GRACEBackbone=GPT-J2026.04 | 35.58 | 6.4 | 0.33 | 100 | |
| BASEBackbone=LLaMA-3 (8B)2026.04 | 33.77 | 0.81 | 0.5 | 100 | |
| BASEBackbone=GPT-J2026.04 | 33.51 | 0.31 | 0.22 | 100 | |
| FTBackbone=GPT-J2026.04 | 15.46 | 32.75 | 11.75 | 1.89 | |
| R-ROMEBackbone=GPT-J2026.04 | 5.1 | 6.4 | 8.9 | 0 | |
| ROMEBackbone=LLaMA-3 (8B)2026.04 | 3 | 4.8 | 4.2 | 0 | |
| FTBackbone=LLaMA-3 (8B)2026.04 | 2.03 | 5.6 | 0.5 | 0 | |
| R-ROMEBackbone=LLaMA-3 (8B)2026.04 | 0.43 | 0.3 | 1 | 0 | |
| MEMITBackbone=LLaMA-3 (8B)2026.04 | 0.13 | 0 | 0 | 0.4 | |
| ROMEBackbone=GPT-J2026.04 | 0.03 | 0.1 | 0 | 0 |