Model Editing on COUNTERFACT
88.2SROME
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| ROMEBase Model=GPT-2 Large (774M)2022.02 | 88.2 | 99.9 | 98.2 | 96.3 | 60.4 | 73.4 | 3.5 | 622.5 | 41.9 | |
| ROMEBase Model=GPT-2 Medium (345M)2022.02 | 87.4 | 100 | 94.9 | 96.4 | 56.9 | 71.8 | 2.8 | 625 | 41.7 | |
| FT+LBase Model=GPT-2 Large (774M)2022.02 | 71.2 | 100 | 96.3 | 63 | 5.1 | 61.5 | 1.1 | 625.2 | 39.3 | |
| FT+LBase Model=GPT-2 Medium (345M)2022.02 | 68 | 100 | 94.9 | 68.5 | 6.1 | 51.3 | -1.7 | 626.1 | 39.3 | |
| GPT-2 MBase Model=GPT-2 Medium (345M)2022.02 | 33.4 | 25 | -3.3 | 27.4 | -3 | 74.9 | 3.6 | 625.8 | 31.4 | |
| GPT-2 LBase Model=GPT-2 Large (774M)2022.02 | 32.8 | 23.9 | -4 | 27.4 | -3.5 | 75.7 | 4.3 | 625.4 | 31.8 |