Machine Unlearning on TOFU forget10 1.0
75.85MUBase
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Base2026.03 | 75.85 | 3.19 | 39.52 | |
| ME_KLTarget Model=Llama-3.1-8B, Method category=Convergence-based2026.03 | 74.44 | 43.03 | 58.73 | |
| IDKVariant=AP2026.03 | 74.24 | 61.27 | 67.76 | |
| MEVariant=KL2026.03 | 73.84 | 44.29 | 59.06 | |
| ASUVariant=GD2026.03 | 73.82 | 78.72 | 76.27 | |
| MEVariant=GD2026.03 | 73.36 | 45.95 | 59.65 | |
| ASUVariant=KL2026.03 | 73.27 | 78.16 | 75.71 | |
| IDK_APTarget Model=Llama-3.1-8B, Method category=Convergence-based2026.03 | 72.7 | 65.79 | 69.24 | |
| ASU_KLTarget Model=Llama-3.1-8B2026.03 | 71.81 | 77 | 74.4 | |
| NPO_KLTarget Model=Llama-3.1-8B, Method category=Divergence-based2026.03 | 65.5 | 57.48 | 61.49 | |
| NPOVariant=KL2026.03 | 57 | 70.37 | 63.68 | |
| NPOVariant=GD2026.03 | 56.58 | 73.04 | 64.81 | |
| GA_KLTarget Model=Llama-3.1-8B, Method category=Divergence-based2026.03 | 54.43 | 1.87 | 28.15 | |
| GAVariant=KL2026.03 | 54.38 | 11.17 | 32.78 | |
| GAVariant=GD2026.03 | 50.29 | 0.01 | 25.15 | |
| IDK_KLTarget Model=Llama-3.1-8B, Method category=Convergence-based2026.03 | 21.72 | 55.77 | 38.75 | |
| DPO_KLTarget Model=Llama-3.1-8B, Method category=Convergence-based2026.03 | 19.5 | 63.58 | 41.54 | |
| IDKVariant=KL2026.03 | 7.52 | 59.06 | 33.29 | |
| IDKVariant=GD2026.03 | 5.26 | 58.8 | 32.03 | |
| DPOVariant=GD2026.03 | 0 | 74.31 | 37.15 | |
| DPOVariant=KL2026.03 | 0 | 74.44 | 37.22 |