LLM Parameter Editing on MCF
99.6Efficacy SuccessFE
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| FEModel=GPT2-XL2026.05 | 99.6 | 98.4 | 93.6 | 60.2 | 71 | 66.9 | 63.2 | 61.7 | |
| AlphaeditModel=GPT2-XL2026.05 | 99.4 | 97.2 | 92.1 | 58.7 | 98 | 62.9 | 57.8 | 55.9 | |
| BLUEModel=GPT2-XL2026.05 | 94.3 | 82.6 | 82.7 | 47.2 | 75 | 67.7 | 62.4 | 60.1 | |
| MEMITModel=GPT2-XL2026.05 | 94 | 79.7 | 82.1 | 45.3 | 81 | 63.5 | 60.8 | 59.4 | |
| PRUNEModel=GPT2-XL2026.05 | 93.8 | 79.6 | 82.2 | 45.3 | 80 | 63.5 | 60.9 | 59.5 | |
| RECTModel=GPT2-XL2026.05 | 93.4 | 78.9 | 81.5 | 44.6 | 80 | 63.7 | 61.1 | 59.7 | |
| OneLayerModel=GPT2-XL2026.05 | 74.4 | 43.3 | 58.7 | 19.6 | 55 | 68.8 | 65.7 | 64.4 | |
| WISEModel=GPT2-XL2026.05 | 24.6 | 0.4 | 26.6 | 0.8 | 21 | 80.5 | 74.9 | 74 |