Sequential Knowledge Editing on CounterFact larger
98.97EfficacyDeltaEdit
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DeltaEditModel=GPT2-XL2025.05 | 98.97 | 91.1 | 92.91 | |
| DeltaEditModel=Llama3-8B2025.05 | 98.63 | 83.2 | 78.67 | |
| AlphaEditModel=GPT2-XL2025.05 | 98.37 | 88.05 | 92.73 | |
| AlphaEditModel=Llama3-8B2025.05 | 93.83 | 84.75 | 64.44 | |
| RECTModel=GPT2-XL2025.05 | 88.3 | 74.8 | 72.85 | |
| FTModel=Llama3-8B2025.05 | 81 | 66.93 | 12.2 | |
| FTModel=GPT2-XL2025.05 | 69.93 | 50.43 | 18.24 | |
| MEMITModel=GPT2-XL2025.05 | 67.8 | 60.85 | 68.99 | |
| ROMEModel=Llama3-8B2025.05 | 63.37 | 59.5 | 24.77 | |
| PRUNEModel=GPT2-XL2025.05 | 60 | 56.13 | 80.94 | |
| ROMEModel=GPT2-XL2025.05 | 53.93 | 49.55 | 84.56 | |
| PRUNEModel=Llama3-8B2025.05 | 51.2 | 51.2 | 17.14 | |
| MEMITModel=Llama3-8B2025.05 | 50.33 | 50.37 | 81.62 | |
| RECTModel=Llama3-8B2025.05 | 49.23 | 49.23 | 16.07 |