Knowledge Editing on FINE-KED GPT2-XL (test)
99.1EfficacyFT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FTBackbone=GPT2-XL (1.5B)2025.05 | 99.1 | 46.8 | 45 | 40.9 | 44.9 | |
| AlphaEditBackbone=GPT2-XL (1.5B)2025.05 | 98.7 | 47.3 | 43.9 | 47.4 | 46.4 | |
| DiKEBackbone=GPT2-XL (1.5B)2025.05 | 97.4 | 54.2 | 51.8 | 60.2 | 55 | |
| MENDBackbone=GPT2-XL (1.5B)2025.05 | 94.2 | 47.1 | 47.5 | 31.5 | 43.4 | |
| ROMEBackbone=GPT2-XL (1.5B)2025.05 | 91.8 | 48.3 | 45.4 | 56.3 | 49.5 | |
| ROME-CBackbone=GPT2-XL (1.5B)2025.05 | 91.7 | 49.7 | 46.7 | 57.4 | 50.8 | |
| MEMIT-CBackbone=GPT2-XL (1.5B)2025.05 | 91 | 49.5 | 45.5 | 55.3 | 49.8 | |
| MEMITBackbone=GPT2-XL (1.5B)2025.05 | 90.8 | 49 | 45 | 55.9 | 49.6 | |
| BASEBackbone=GPT2-XL (1.5B)2025.05 | 20.4 | 62.7 | 65.7 | 71.9 | 65.8 |