Machine Unlearning on MUSE-Books (test)
99.702No VerbMemOriginal
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Original2025.09 | 99.702 | 45.879 | -57.148 | 69.4 | — | — | — | — | — | — | — | — | — | |
| Task VectorRuntime (min)=68.4172025.09 | 99.701 | 45.879 | -57.148 | 69.4 | 0 | 0 | 1 | 0 | 7 | — | — | — | — | |
| DOORRuntime (min)=300.0672025.09 | 99.69 | 33.115 | -54.979 | 57.105 | 0.088 | 0.067 | 0.823 | 0.169 | 5 | — | — | — | — | |
| TargetBackbone=Llama-3.1-8B2025.06 | 65 | 62 | -100 | — | — | — | — | — | — | 0.64 | 0.96 | 1 | 1 | |
| TargetDescription=Finetuned on all MUSE books2025.06 | 49 | 62 | -100 | — | — | — | — | — | — | 0.43 | 0.94 | 1 | 1 | |
| SatImp2025.06 | 43 | 62 | -100 | — | — | — | — | — | — | 0.37 | 0.93 | 1 | 1 | |
| SatImpBackbone=Llama-3.1-8B, gamma=42025.06 | 43 | 54 | -100 | — | — | — | — | — | — | 0.39 | 0.95 | 1 | 1 | |
| IdealDescription=Reference model2025.06 | 17 | 67 | 0 | — | — | — | — | — | — | 0.02 | 0.54 | 0.45 | 0.39 | |
| UNDIALBackbone=Llama-3.1-8B, lambda=42025.06 | 17 | 35 | -98.15 | — | — | — | — | — | — | 0.02 | 0.68 | 0.99 | 0.99 | |
| MSACheckpoint=last2025.06 | 16 | 65 | 14.67 | — | — | — | — | — | — | 0.02 | 0.55 | 0.37 | 0.22 | |
| UNDIAL2025.06 | 16 | 53 | -100 | — | — | — | — | — | — | 0.02 | 0.64 | 1 | 1 | |
| IdealBackbone=Llama-3.1-8B2025.06 | 16 | 64 | 0 | — | — | — | — | — | — | 0.02 | 0.52 | 0.51 | 0.47 | |
| MSACheckpoint=3691B2025.06 | 15 | 63 | 27.63 | — | — | — | — | — | — | 0.02 | 0.51 | 0.3 | 0.21 | |
| MSACheckpoint=3859B2025.06 | 15 | 59 | 23.45 | — | — | — | — | — | — | 0.02 | 0.51 | 0.23 | 0.16 | |
| NPO2025.06 | 15 | 62 | -99.93 | — | — | — | — | — | — | 0.02 | 0.64 | 1 | 0.99 | |
| NPOBackbone=Llama-3.1-8B, lambda=42025.06 | 14 | 58 | -99.9 | — | — | — | — | — | — | 0.02 | 0.58 | 1 | 0.84 | |
| Task-Vector2025.06 | 13 | 48 | -84.3 | — | — | — | — | — | — | 0.01 | 0.46 | 0.92 | 0.95 | |
| MSAbaseBackbone=Llama-3.1-8B2025.06 | 13 | 55 | -1.37 | — | — | — | — | — | — | 0.01 | 0.48 | 0.52 | 0.52 | |
| MSACheckpoint=500B2025.06 | 12 | 51 | 56.38 | — | — | — | — | — | — | 0.01 | 0.41 | 0.14 | 0.09 | |
| MSACheckpoint=2207B2025.06 | 10 | 45 | 74.05 | — | — | — | — | — | — | 0.01 | 0.37 | 0.04 | 0.01 | |
| RMU2025.06 | 8 | 64 | -17.83 | — | — | — | — | — | — | 0.01 | 0.06 | 0.55 | 0.47 | |
| GradDiff2025.06 | 1 | 45 | -9.47 | — | — | — | — | — | — | 0.01 | 0.2 | 0.5 | 0.45 | |
| RMUBackbone=Llama-3.1-8B, lambda=42025.06 | 1 | 52 | -46.44 | — | — | — | — | — | — | 0.01 | 0.04 | 0.74 | 0.62 | |
| GradDiffBackbone=Llama-3.1-8B, lambda=42025.06 | 1 | 21 | 38.06 | — | — | — | — | — | — | 0.01 | 0.01 | 0.32 | 0.49 | |
| SAM+NPORuntime (min)=25.6112025.09 | 0 | 0 | -28.536 | 32.766 | 1 | 0.886 | 0.472 | 0.748 | 2 | — | — | — | — | |
| NPORuntime (min)=15.0012025.09 | 0 | 0 | -30.817 | 31.273 | 1 | 0.816 | 0.451 | 0.717 | 3 | — | — | — | — | |
| NPOGDRRuntime (min)=17.9172025.09 | 0 | 0 | -30.966 | 30.291 | 1 | 0.811 | 0.437 | 0.708 | 4 | — | — | — | — | |
| GARuntime (min)=10.0732025.09 | 0 | 0 | -27.831 | 0 | 1 | 0.908 | 0 | 0 | 7 | — | — | — | — | |
| GAGDRRuntime (min)=15.3482025.09 | 0 | 0 | -24.866 | 0.207 | 1 | 1 | 0.003 | 0.144 | 6 | — | — | — | — | |
| PRISMRuntime (min)=25.9992025.09 | 0 | 0 | -28.39 | 49.616 | 1 | 0.891 | 0.715 | 0.86 | 1 | — | — | — | — |