List your company
ResearchFormerly WizWand

Machine Unlearning on WMDP (Unlearn Score)

76.1Unlearn ScorePRISM
-3.04417.50338.0558.597Sep 27, 2025
Updated 7mo ago

Evaluation results

MethodLinks
PRISMBase Model=Mistral-8B-Instruct-24102025.0976.1
SAM+NPOBase Model=Mistral-8B-Instruct-24102025.0972.1
NPOGDRBase Model=Mistral-8B-Instruct-24102025.0955.6
GAGDRBase Model=Mistral-8B-Instruct-24102025.0952.8
PRISMBase Model=Llama-2 7B2025.0952.1
NPOGDRBase Model=Llama-2 7B2025.0951.9
DOORBase Model=Llama-2 7B2025.0947.9
GAGDRBase Model=Llama-2 7B2025.0946.9
SAM+NPOBase Model=Llama-2 7B2025.0944.3
NPOBase Model=Llama-2 7B2025.0932.2
DOORBase Model=Mistral-8B-Instruct-24102025.0928.9
NPOBase Model=Mistral-8B-Instruct-24102025.095.5
GABase Model=Llama-2 7B2025.090
GABase Model=Mistral-8B-Instruct-24102025.090
Task VectorBase Model=Llama-2 7B2025.090
Task VectorBase Model=Mistral-8B-Instruct-24102025.090