List your company
ResearchFormerly WizWand

Toxicity Unlearning on Toxicity

0.45S-unlearning Scoremulti-layer toxicity unlearning objective
0.28360.32680.370.4132May 26, 2026
Updated 5mo ago

Evaluation results

MethodLinks
multi-layer toxicity unlearning objectiveBase Model=Qwen2.5-7B2026.050.45
multi-layer toxicity unlearning objectiveBase Model=Llama3.1-8B2026.050.43
BaseBase Model=Qwen2.5-7B2026.050.43
RMUBase Model=Qwen2.5-7B2026.050.43
Adaptive RMUBase Model=Qwen2.5-7B2026.050.41
multi-layer toxicity unlearning objectiveBase Model=Olmo3-7B2026.050.35
BaseBase Model=Llama3.1-8B2026.050.34
multi-layer toxicity unlearning objectiveBase Model=Mistral-7B2026.050.34
RMUBase Model=Llama3.1-8B2026.050.33
Adaptive RMUBase Model=Llama3.1-8B2026.050.32
BaseBase Model=Mistral-7B2026.050.32
BaseBase Model=Olmo3-7B2026.050.32
RMUBase Model=Mistral-7B2026.050.31
RMUBase Model=Olmo3-7B2026.050.3
Adaptive RMUBase Model=Mistral-7B2026.050.29
Adaptive RMUBase Model=Olmo3-7B2026.050.29