Lifelong Knowledge Editing on GPT-2 XL (1024 Edits, test)
19.3Score (S)WilKE
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| WilKEBackbone=GPT-2 XL (1.5B)2024.02 | 19.3 | 7,070 | 5,100 | 1,270 | 1,610 | 1,310 | |
| MEMITBackbone=GPT-2 XL (1.5B)2024.02 | 13.2 | 9,250 | 5,510 | 3,500 | 660 | 660 | |
| ROMEBackbone=GPT-2 XL (1.5B)2024.02 | 9.3 | 1,580 | 880 | 680 | 1,220 | 790 | |
| KEBackbone=GPT-2 XL (1.5B)2024.02 | 0 | 10 | 10 | 0 | 0 | 0 | |
| KNBackbone=GPT-2 XL (1.5B)2024.02 | 0 | 10 | 100 | 0 | 0 | 0 | |
| MENDBackbone=GPT-2 XL (1.5B)2024.02 | 0 | 50 | 10 | 40 | 0 | 0 |