Knowledge Editing on Counterfact 10,000 facts
10,000Relational ScoreGRACE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRACEModel=GPT-XL2025.03 | 10,000 | 39 | 6,875 | 5,638 | |
| GRACEModel=GPT-J2025.03 | 10,000 | 29 | 9,908 | 6,646 | |
| GRACEModel=LLaMA32025.03 | 10,000 | 533 | 10,000 | 6,844 | |
| MEMITModel=GPT-J2025.03 | 9,580 | 5,764 | 6,187 | 7,177 | |
| MindBridgeModel=GPT-XL2025.03 | 9,560 | 8,110 | 3,818 | 7,162 | |
| MindBridgeModel=GPT-J2025.03 | 9,460 | 8,260 | 9,356 | 9,025 | |
| MindBridgeModel=LLaMA32025.03 | 9,385 | 8,335 | 9,214 | 8,978 | |
| AlphaEditModel=GPT-J2025.03 | 9,200 | 4,645 | 5,740 | 6,528 | |
| AlphaEditModel=LLaMA32025.03 | 9,170 | 5,154 | 5,558 | 6,628 | |
| EMMETModel=GPT-XL2025.03 | 8,551 | 4,665 | 6,217 | 6,478 | |
| EMMETModel=GPT-J2025.03 | 8,551 | 4,665 | 6,217 | 6,478 | |
| MEMITModel=LLaMA32025.03 | 8,471 | 4,120 | 6,449 | 6,347 | |
| AlphaEditModel=GPT-XL2025.03 | 6,600 | 3,090 | 4,298 | 4,662 | |
| EMMETModel=LLaMA32025.03 | 6,053 | 3,181 | 3,228 | 4,154 | |
| MEMITModel=GPT-XL2025.03 | 4,536 | 3,970 | 6,625 | 4,408 | |
| FT-LModel=LLaMA32025.03 | 2,567 | 924 | 23 | 1,171 | |
| WISEModel=GPT-J2025.03 | 1,838 | 1,208 | 385 | 1,144 | |
| WISEModel=LLaMA32025.03 | 1,608 | 1,068 | 387 | 1,021 | |
| FT-LModel=GPT-J2025.03 | 1,058 | 739 | 103 | 634 | |
| WISEModel=GPT-XL2025.03 | 179 | 199 | 5,038 | 1,806 | |
| Pre-editedModel=LLaMA32025.03 | 70 | 130 | — | — | |
| Pre-editedModel=GPT-XL2025.03 | 28 | 42 | — | — | |
| Pre-editedModel=GPT-J2025.03 | 9 | 29 | — | — | |
| r-ROMEModel=LLaMA32025.03 | 9 | 0 | 15 | 8 | |
| FT-LModel=GPT-XL2025.03 | 0 | 0 | 2,284 | 761 | |
| r-ROMEModel=GPT-XL2025.03 | 0 | 0 | 0 | 0 | |
| r-ROMEModel=GPT-J2025.03 | 0 | 0 | 0 | 0 |