List your company

Safety Unlearning on BeaverTails

0Harmful Response ProbabilityRepSelect
-0.8084.64610.115.554Jun 15, 2026

Evaluation results

MethodLinks
RepSelectModel=Llama-3.1-8B, Attack=No attack2026.060
GradDiffModel=Llama-3.1-8B, Attack=No attack2026.060
RepSelectModel=Qwen3.5-9B, Attack=No attack2026.060
RepSelectModel=Qwen3.5-9B, Attack=Few-shot k=52026.060
RepSelectModel=Qwen3.5-9B, Attack=Few-shot k=102026.060
GradDiffModel=Qwen3.5-9B, Attack=No attack2026.060.3
RMUModel=Llama-3.1-8B, Attack=No attack2026.060.4
RMUModel=Qwen3.5-9B, Attack=No attack2026.060.8
RMUModel=Qwen3.5-9B, Attack=Few-shot k=52026.060.9
RepSelectModel=Llama-3.1-8B, Attack=Few-shot k=102026.061
RepSelectModel=Llama-3.1-8B, Attack=Few-shot k=52026.061.3
RMUModel=Qwen3.5-9B, Attack=Few-shot k=102026.061.5
GradDiffModel=Llama-3.1-8B, Attack=Few-shot k=52026.062
SimNPOModel=Llama-3.1-8B, Attack=No attack2026.062.7
GradDiffModel=Llama-3.1-8B, Attack=Few-shot k=102026.063
SimNPOModel=Llama-3.1-8B, Attack=Few-shot k=52026.063.1
SimNPOModel=Llama-3.1-8B, Attack=Few-shot k=102026.063.1
GradDiffModel=Qwen3.5-9B, Attack=Few-shot k=52026.063.6
UNDIALModel=Qwen3.5-9B, Attack=No attack2026.064.6
GradDiffModel=Qwen3.5-9B, Attack=Few-shot k=102026.067.3
NPOModel=Qwen3.5-9B, Attack=No attack2026.067.8
NPOModel=Qwen3.5-9B, Attack=Few-shot k=52026.067.9
NPOModel=Qwen3.5-9B, Attack=Few-shot k=102026.068.5
NPOModel=Llama-3.1-8B, Attack=No attack2026.0611.9
No unlearnModel=Qwen3.5-9B, Attack=No attack2026.0612.8
No unlearnModel=Qwen3.5-9B, Attack=Few-shot k=52026.0614.4
NPOModel=Llama-3.1-8B, Attack=Few-shot k=52026.0614.5
UNDIALModel=Qwen3.5-9B, Attack=Few-shot k=52026.0614.5
No unlearnModel=Qwen3.5-9B, Attack=Few-shot k=102026.0614.8
UNDIALModel=Qwen3.5-9B, Attack=Few-shot k=102026.0614.9
NPOModel=Llama-3.1-8B, Attack=Few-shot k=102026.0615.1
UNDIALModel=Llama-3.1-8B, Attack=No attack2026.0615.6
No unlearnModel=Llama-3.1-8B, Attack=No attack2026.0615.6
SimNPOModel=Qwen3.5-9B, Attack=No attack2026.0616.4
SimNPOModel=Qwen3.5-9B, Attack=Few-shot k=52026.0617.6
RMUModel=Llama-3.1-8B, Attack=Few-shot k=52026.0617.8
SimNPOModel=Qwen3.5-9B, Attack=Few-shot k=102026.0617.8
RMUModel=Llama-3.1-8B, Attack=Few-shot k=102026.0618.6
UNDIALModel=Llama-3.1-8B, Attack=Few-shot k=52026.0619.9
No unlearnModel=Llama-3.1-8B, Attack=Few-shot k=52026.0620
UNDIALModel=Llama-3.1-8B, Attack=Few-shot k=102026.0620.2
No unlearnModel=Llama-3.1-8B, Attack=Few-shot k=102026.0620.2