Knowledge Editing on Popular dataset
50CISAKE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SAKEBackbone Model=GPT2-XL2025.03 | 50 | 33.33 | 54.59 | 58.39 | |
| SAKEBackbone Model=Llama 2-7b2025.03 | 44.32 | 27.63 | 53.34 | 56.93 | |
| ROMEBackbone Model=GPT2-XL2025.03 | 38.62 | 16.67 | 51.96 | 39.43 | |
| ROMEBackbone Model=Llama 2-7b2025.03 | 27.51 | 8.28 | 47.72 | 29.9 | |
| ActAddBackbone Model=GPT2-XL2025.03 | 26.63 | 29.17 | 42.12 | 50.68 | |
| CompMEMITBackbone Model=GPT2-XL2025.03 | 20.69 | 0 | 10.1 | 14.71 | |
| CompMEMITBackbone Model=Llama 2-7b2025.03 | 17.21 | 7.89 | 8.45 | 48.83 | |
| ActAddBackbone Model=Llama 2-7b2025.03 | 9.57 | 18.12 | 47.69 | 59.4 | |
| MEMITBackbone Model=Llama 2-7b2025.03 | 5.8 | 9.28 | 42.43 | 24.52 | |
| MEMITBackbone Model=GPT2-XL2025.03 | 2.47 | 1.95 | 7.17 | 3.75 |