Machine Unlearning on WMDP (Unlearn Score)
76.1Unlearn ScorePRISM
Evaluation Results
| Method | Links | |
|---|---|---|
| PRISMBase Model=Mistral-8B-Instruct-24102025.09 | 76.1 | |
| SAM+NPOBase Model=Mistral-8B-Instruct-24102025.09 | 72.1 | |
| NPOGDRBase Model=Mistral-8B-Instruct-24102025.09 | 55.6 | |
| GAGDRBase Model=Mistral-8B-Instruct-24102025.09 | 52.8 | |
| PRISMBase Model=Llama-2 7B2025.09 | 52.1 | |
| NPOGDRBase Model=Llama-2 7B2025.09 | 51.9 | |
| DOORBase Model=Llama-2 7B2025.09 | 47.9 | |
| GAGDRBase Model=Llama-2 7B2025.09 | 46.9 | |
| SAM+NPOBase Model=Llama-2 7B2025.09 | 44.3 | |
| NPOBase Model=Llama-2 7B2025.09 | 32.2 | |
| DOORBase Model=Mistral-8B-Instruct-24102025.09 | 28.9 | |
| NPOBase Model=Mistral-8B-Instruct-24102025.09 | 5.5 | |
| GABase Model=Llama-2 7B2025.09 | 0 | |
| GABase Model=Mistral-8B-Instruct-24102025.09 | 0 | |
| Task VectorBase Model=Llama-2 7B2025.09 | 0 | |
| Task VectorBase Model=Mistral-8B-Instruct-24102025.09 | 0 |