Machine Unlearning on TOFU 1.0 (Retain Set)
100ROUGE-LMISTRAL-7B-TOFU
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MISTRAL-7B-TOFUModel Type=Base (No Unlearning)2024.10 | 100 | 1 | 48 | |
| LLAMA2-13B-TOFUModel Type=Base (No Unlearning)2024.10 | 99 | 1 | 53 | |
| CMCBase Model=MISTRAL-7B-TOFU2024.10 | 99 | 0.99 | 51 | |
| WISECategory=Model Editing, Target=Dummy2025.12 | 98.2 | 0.803 | 46.8 | |
| WISECategory=Model Editing, Target=Incorrect2025.12 | 98.2 | 0.918 | 42.5 | |
| WISECategory=Model Editing, Target=Avoidant2025.12 | 98.2 | 0.949 | 48.7 | |
| LLAMA2-7B-TOFUModel Type=Base (No Unlearning)2024.10 | 98 | 0.99 | 48 | |
| Ground Truth2025.12 | 97.6 | 0.989 | 47.1 | |
| CMCBase Model=LLAMA2-13B-TOFU2024.10 | 97 | 0.99 | 55 | |
| BaseBackbone=Dream-7B-Instruct2026.05 | 96.6 | 0.891 | — | |
| LoRABase Model=MISTRAL-7B-TOFU2024.10 | 95 | 0.96 | 49 | |
| MDUBackbone=Dream-7B-Instruct, τ=0.502026.05 | 93.1 | 0.795 | — | |
| CMCBase Model=LLAMA2-7B-TOFU2024.10 | 91 | 0.97 | 51 | |
| MDUBackbone=Dream-7B-Instruct, τ=0.002026.05 | 90.4 | 0.672 | — | |
| MDUBackbone=Dream-7B-Instruct, τ=1.002026.05 | 90.3 | 0.694 | — | |
| MDUBackbone=Dream-7B-Instruct, τ=0.252026.05 | 89.9 | 0.758 | — | |
| BaseBackbone=LLaDA-8B-Instruct2026.05 | 87 | 0.33 | — | |
| MDUBackbone=LLaDA-8B-Instruct, τ=0.002026.05 | 86.8 | 0.381 | — | |
| MDUBackbone=LLaDA-8B-Instruct, τ=0.252026.05 | 85.7 | 0.392 | — | |
| MDUBackbone=LLaDA-8B-Instruct, τ=0.502026.05 | 85.3 | 0.447 | — | |
| SimNPOBackbone=Dream-7B-Instruct2026.05 | 85.3 | 0.625 | — | |
| DPOBackbone=Dream-7B-Instruct2026.05 | 85.2 | 0.765 | — | |
| LoRABase Model=LLAMA2-13B-TOFU2024.10 | 85 | 0.92 | 52 | |
| MDUBackbone=Dream-7B-Instruct, τ=0.752026.05 | 84.2 | 0.437 | — | |
| WGABackbone=Dream-7B-Instruct2026.05 | 82.1 | 0.62 | — | |
| SimNPOBackbone=LLaDA-8B-Instruct2026.05 | 80.4 | 0.273 | — | |
| DPOBackbone=LLaDA-8B-Instruct2026.05 | 79.6 | 0.401 | — | |
| GDBackbone=Dream-7B-Instruct2026.05 | 78.3 | 0.576 | — | |
| POCategory=Model Unlearning2025.12 | 77.3 | 0.942 | 45.5 | |
| NPOBackbone=Dream-7B-Instruct2026.05 | 74.2 | 0.528 | — | |
| NPOBackbone=LLaDA-8B-Instruct2026.05 | 72.6 | 0.138 | — | |
| LoRABase Model=LLAMA2-7B-TOFU2024.10 | 71 | 0.75 | 48 | |
| WGABackbone=LLaDA-8B-Instruct2026.05 | 69.6 | 0.304 | — | |
| MDUBackbone=LLaDA-8B-Instruct, τ=0.752026.05 | 68.4 | 0.535 | — | |
| GDBackbone=LLaDA-8B-Instruct2026.05 | 67.6 | 0.169 | — | |
| δ-UNLEARNINGBase Model=LLAMA2-13B-TOFU2024.10 | 53 | 0.48 | 52 | |
| MDUBackbone=LLaDA-8B-Instruct, τ=1.002026.05 | 51.1 | 0.485 | — | |
| GABackbone=Dream-7B-Instruct2026.05 | 48.7 | 0.111 | — | |
| GDCategory=Model Unlearning2025.12 | 46.6 | 0.548 | 49.3 | |
| GABackbone=LLaDA-8B-Instruct2026.05 | 36.1 | 0.018 | — | |
| IKECategory=Model Editing, Target=Incorrect2025.12 | 29 | 0.781 | 57.4 | |
| IKECategory=Model Editing, Target=Dummy2025.12 | 28.3 | 0.713 | 58.3 | |
| IKECategory=Model Editing, Target=Avoidant2025.12 | 25.1 | 0.784 | 58.3 | |
| ROMECategory=Model Editing, Target=Incorrect2025.12 | 9 | — | — | |
| ROMECategory=Model Editing, Target=Dummy2025.12 | 8.9 | 0.041 | 22.8 | |
| ROMECategory=Model Editing, Target=Avoidant2025.12 | 3 | — | — | |
| KLCategory=Model Unlearning2025.12 | 0 | 0 | 6.7 | |
| GACategory=Model Unlearning2025.12 | 0 | 0 | 7.5 |