Safety Evaluation on WildGuard (Safe)
30.7Refusal RateTarget feature
Evaluation Results
| Method | Links | |
|---|---|---|
| Target featureModel=Qwen3-4B-it2026.06 | 30.7 | |
| Target featureModel=Qwen3-1.7B-it2026.06 | 18.3 | |
| Random featureModel=Qwen3-1.7B-it2026.06 | 5.3 | |
| No SteeringModel=Qwen3-4B-it2026.06 | 4.4 | |
| No SteeringModel=Qwen3-1.7B-it2026.06 | 4.3 | |
| Random featureModel=Qwen3-4B-it2026.06 | 3.6 |