Knowledge Editing on EVK-Bench
100Embedding Stability (ES)Pre-edited
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Pre-editedModel=GPT2-XL2026.02 | 100 | 100 | |
| Pre-editedModel=LLaMA32026.02 | 100 | 100 | |
| Pre-editedModel=GPT-J2026.02 | 100 | 100 | |
| EVK-EditModel=GPT-J2026.02 | 88.15 | 83.75 | |
| AlphaEditModel=GPT-J2026.02 | 87.13 | 83.36 | |
| EVK-EditModel=LLaMA32026.02 | 84.43 | 74.74 | |
| RECTModel=GPT-J2026.02 | 84.09 | 77.22 | |
| RECTModel=LLaMA32026.02 | 83.77 | 72.42 | |
| MEMITModel=GPT-J2026.02 | 83.73 | 75.7 | |
| MEMITModel=LLaMA32026.02 | 83.67 | 71.4 | |
| AlphaEditModel=LLaMA32026.02 | 83.3 | 73.98 | |
| PRUNEModel=LLaMA32026.02 | 74.1 | 64.63 | |
| PRUNEModel=GPT-J2026.02 | 71.91 | 62.24 | |
| EVK-EditModel=GPT2-XL2026.02 | 69.52 | 76.6 | |
| AlphaEditModel=GPT2-XL2026.02 | 67.7 | 75.58 | |
| RECTModel=GPT2-XL2026.02 | 67.57 | 69.46 | |
| MEMITModel=GPT2-XL2026.02 | 65.26 | 67.92 | |
| PRUNEModel=GPT2-XL2026.02 | 59.74 | 61.34 | |
| ROMEModel=GPT-J2026.02 | 55.45 | 57.27 | |
| ROMEModel=GPT2-XL2026.02 | 54.08 | 56.92 | |
| ROMEModel=LLaMA32026.02 | 52.08 | 50.56 |