Knowledge Editing on ZsRE 10,000 facts
100ReliabilityGRACE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRACEModel=GPT-XL2025.03 | 100 | 3.17 | 100 | 67.72 | |
| GRACEModel=GPT-J2025.03 | 100 | 3.13 | 100 | 67.71 | |
| GRACEModel=LLaMA32025.03 | 100 | 5.33 | 100 | 68.44 | |
| MindBridgeModel=GPT-J2025.03 | 99.06 | 83.71 | 95.77 | 92.85 | |
| MindBridgeModel=LLaMA32025.03 | 99.03 | 85.5 | 92.65 | 92.39 | |
| MEMITModel=GPT-J2025.03 | 86.17 | 69.7 | 75.96 | 77.28 | |
| AlphaEditModel=LLaMA32025.03 | 78.64 | 62.06 | 73.57 | 71.43 | |
| MindBridgeModel=GPT-XL2025.03 | 78.14 | 68.17 | 85.67 | 77.33 | |
| AlphaEditModel=GPT-J2025.03 | 76.95 | 52.37 | 62.02 | 63.78 | |
| EMMETModel=GPT-XL2025.03 | 70.37 | 60.34 | 78.37 | 69.69 | |
| EMMETModel=GPT-J2025.03 | 70.37 | 60.34 | 78.37 | 69.69 | |
| MEMITModel=LLaMA32025.03 | 66.16 | 59.92 | 80.51 | 68.86 | |
| EMMETModel=LLaMA32025.03 | 62.94 | 59.64 | 31.95 | 51.51 | |
| AlphaEditModel=GPT-XL2025.03 | 51.82 | 42.97 | 53.72 | 49.5 | |
| MEMITModel=GPT-XL2025.03 | 47.2 | 24.7 | 60.34 | 50.44 | |
| WISEModel=GPT-J2025.03 | 36.88 | 34.53 | 99.51 | 56.97 | |
| WISEModel=LLaMA32025.03 | 29.18 | 29.18 | 99.4 | 52.07 | |
| Pre-editedModel=LLaMA32025.03 | 27.7 | 27.08 | — | — | |
| WISEModel=GPT-XL2025.03 | 24.88 | 24.84 | 99.98 | 49.9 | |
| Pre-editedModel=GPT-J2025.03 | 23.01 | 22.25 | — | — | |
| Pre-editedModel=GPT-XL2025.03 | 22.44 | 21.56 | — | — | |
| FT-LModel=GPT-J2025.03 | 15.73 | 14.04 | 9.47 | 13.08 | |
| FT-LModel=GPT-XL2025.03 | 15.6 | 15.83 | 28.21 | 19.88 | |
| FT-LModel=LLaMA32025.03 | 5.99 | 3.92 | 0.64 | 3.52 | |
| r-ROMEModel=LLaMA32025.03 | 0.7 | 0.5 | 1.01 | 0.74 | |
| r-ROMEModel=GPT-J2025.03 | 0.09 | 0.08 | 0 | 0.05 | |
| r-ROMEModel=GPT-XL2025.03 | 0 | 0 | 0 | 0 |