Knowledge Editing on ZsRE 100 edits Sequential (test)
100EfficiencyGRACE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRACEModel=GPT2024.10 | 100 | 0.4 | 100 | 66.8 | |
| GRACEModel=Llama2024.10 | 99.33 | 0.53 | 99.88 | 66.58 | |
| MEMITModel=GPT2024.10 | 98.25 | 93.41 | 87.24 | 92.97 | |
| D4SModel=GPT2024.10 | 97.72 | 97.01 | 85.3 | 93.34 | |
| ROMEModel=Llama2024.10 | 97.68 | 92.39 | 90.22 | 93.43 | |
| D4SModel=Llama2024.10 | 95.53 | 86.45 | 93.44 | 91.81 | |
| MEMITModel=Llama2024.10 | 92.99 | 83.22 | 94.02 | 90.08 | |
| FTModel=Llama2024.10 | 30.47 | 25.78 | 12.19 | 22.81 | |
| FTModel=GPT2024.10 | 15.28 | 7.4 | 0.42 | 7.7 | |
| ROMEModel=GPT2024.10 | 1.17 | 1.5 | 0.06 | 0.91 | |
| PMETModel=GPT2024.10 | 0.33 | 0.33 | 0 | 0.22 | |
| PMETModel=Llama2024.10 | 0.24 | 0.35 | 0.46 | 0.35 |