List your company

Safety Evaluation on Alpaca Mix (Unsafe)

93Refusal RateTarget feature
-3.40821.62146.6571.679Jun 25, 2026

Evaluation results

MethodLinks
Target featureModel=Qwen3-4B-it2026.0693
No SteeringModel=Qwen3-4B-it2026.0676
Target featureModel=Qwen3-1.7B-it2026.0666.7
No SteeringModel=Qwen3-1.7B-it2026.0641.7
Random featureModel=Qwen3-4B-it2026.061.2
Random featureModel=Qwen3-1.7B-it2026.060.3