LLM Parameter Editing on MCF (test)
99.9Efficacy SuccessFE
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| FEModel=GPT-J-6B2026.05 | 99.9 | 99.4 | 95.4 | 65.7 | 0.42 | 78.4 | 73.6 | 72.8 | |
| MEMITModel=GPT-J-6B2026.05 | 99.8 | 99.2 | 91.8 | 61.9 | 0.44 | 75 | 71.2 | 70.8 | |
| BLUEModel=GPT-J-6B2026.05 | 99.8 | 99.1 | 94.9 | 69.2 | 0.38 | 78 | 72.9 | 72.4 | |
| PRUNEModel=GPT-J-6B2026.05 | 99.8 | 99.3 | 91.8 | 61.8 | 0.43 | 75.1 | 71.2 | 70.9 | |
| RECTModel=GPT-J-6B2026.05 | 99.8 | 99.2 | 91.5 | 61.2 | 0.43 | 75.1 | 71.3 | 71 | |
| AlphaeditModel=GPT-J-6B2026.05 | 99.8 | 98.6 | 93.2 | 61.5 | 0.58 | 72.5 | 66 | 65.1 | |
| OneLayerModel=GPT-J-6B2026.05 | 96.6 | 85.4 | 79.1 | 41 | 0.42 | 73.4 | 69.1 | 68.7 | |
| WISEModel=GPT-J-6B2026.05 | 27.3 | 6.8 | 50.9 | 15.2 | 4.47 | 45.8 | 48 | 49.5 |