Machine Unlearning on TOFU Llama3.1-8B (forget10)
98Forgetting Quality (FQ)JENSUN
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| JENSUNBase Model=Llama-3.1-8B-Instruct2026.06 | 98 | — | — | — | — | — | — | — | — | — | 9.6 | 88.3 | 65.3 | 236 | 11 | |
| ATWUBase Model=Llama-3.1-8B-Instruct2026.06 | 95.2 | — | — | — | — | — | — | — | — | — | 3.5 | 91.7 | 66.5 | 558 | 58 | |
| RMUBase Model=Llama-3.1-8B-Instruct2026.06 | 87.2 | — | — | — | — | — | — | — | — | — | 1.3 | 85.9 | 65.3 | 540 | 50.5 | |
| SATIMPBase Model=Llama-3.1-8B-Instruct2026.06 | 78.6 | — | — | — | — | — | — | — | — | — | 3 | 75.6 | 65.9 | 549 | 53.5 | |
| SIMNPOBase Model=Llama-3.1-8B-Instruct2026.06 | 69.3 | — | — | — | — | — | — | — | — | — | 6.5 | 62.8 | 65.8 | 541 | 50.5 | |
| WGABase Model=Llama-3.1-8B-Instruct2026.06 | 66.8 | — | — | — | — | — | — | — | — | — | 4.8 | 62.1 | 66 | 550 | 60 | |
| DPOBase Model=Llama-3.1-8B-Instruct2026.06 | 59.5 | — | — | — | — | — | — | — | — | — | 15 | 44.5 | 64.3 | 533 | 38.5 | |
| NPOBase Model=Llama-3.1-8B-Instruct2026.06 | 48 | — | — | — | — | — | — | — | — | — | 5 | 43 | 66.2 | 551 | 48.5 | |
| GRADDIFFBase Model=Llama-3.1-8B-Instruct2026.06 | 39.7 | — | — | — | — | — | — | — | — | — | 17.3 | 22.4 | 66.1 | 545 | 48 | |
| GD+SineLoRA Rank=r=4, Base Model=Llama3.1-8B2025.09 | 0.83 | 0.31 | 0.5 | 0.93 | 0.48 | 0.94 | 0.62 | 0.89 | 0.6 | 0.68 | — | — | — | — | — | |
| GD+SineLoRA Rank=r=8, Base Model=Llama3.1-8B2025.09 | 0.8 | 0.31 | 0.5 | 0.93 | 0.48 | 0.94 | 0.62 | 0.89 | 0.6 | 0.68 | — | — | — | — | — | |
| GD+SineLoRA Rank=r=16, Base Model=Llama3.1-8B2025.09 | 0.8 | 0.31 | 0.5 | 0.93 | 0.48 | 0.94 | 0.62 | 0.89 | 0.6 | 0.68 | — | — | — | — | — | |
| GD+SineLoRA Rank=r=32, Base Model=Llama3.1-8B2025.09 | 0.8 | 0.31 | 0.5 | 0.93 | 0.48 | 0.94 | 0.62 | 0.89 | 0.6 | 0.68 | — | — | — | — | — | |
| ORIGINALBase Model=Llama-3.1-8B-Instruct2026.06 | 0 | — | — | — | — | — | — | — | — | — | 0 | 0 | 66.6 | 546 | 50 |