Sequential Knowledge Editing on CounterFact
100EfficacyGRACE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRACEModel=GPT2024.10 | 100 | 0.5 | 100 | 66.83 | |
| GRACEModel=Llama2024.10 | 99.9 | 0.25 | 99.97 | 66.71 | |
| D4SModel=GPT2024.10 | 99.1 | 47 | 63.7 | 69.93 | |
| D4SModel=Llama2024.10 | 96.68 | 46.66 | 72.45 | 71.93 | |
| MEMITModel=GPT2024.10 | 86.2 | 59.5 | 31.3 | 59 | |
| FTModel=GPT2024.10 | 32.8 | 9 | 1 | 14.27 | |
| ROMEModel=Llama2024.10 | 27.83 | 16.03 | 5.66 | 16.5 | |
| FTModel=Llama2024.10 | 8.46 | 4.07 | 2.03 | 4.85 | |
| ROMEModel=GPT2024.10 | 0.6 | 0.7 | 0.6 | 0.63 | |
| MEMITModel=Llama2024.10 | 0 | 0 | 6.72 | 2.24 |