Knowledge Editing on ZsRE 1000 edits Sequential (test)
100EfficiencyGRACE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRACEModel=GPT2024.10 | 100 | 0.07 | 100 | 66.69 | |
| GRACEModel=Llama2024.10 | 98.94 | 0.35 | 99.77 | 66.35 | |
| D4SModel=GPT2024.10 | 95.98 | 91.17 | 70.17 | 88.77 | |
| D4SModel=Llama2024.10 | 87.36 | 79.3 | 78.99 | 81.88 | |
| ROMEModel=Llama2024.10 | 21.84 | 20.74 | 4.01 | 15.53 | |
| FTModel=Llama2024.10 | 16.13 | 13.89 | 5.96 | 12 | |
| FTModel=GPT2024.10 | 7.03 | 4.15 | 0.11 | 3.77 | |
| MEMITModel=Llama2024.10 | 2.85 | 2.85 | 2.74 | 2.82 | |
| ROMEModel=GPT2024.10 | 2.44 | 2.23 | 0.92 | 1.86 | |
| PMETModel=Llama2024.10 | 0 | 0 | 0 | 0 | |
| MEMITModel=GPT2024.10 | 0 | 0.02 | 0 | 0.01 | |
| PMETModel=GPT2024.10 | 0 | 0 | 0 | 0 |