Safety Evaluation on Alpaca Mix (Unsafe)
93Refusal RateTarget feature
Evaluation Results
| Method | Links | |
|---|---|---|
| Target featureModel=Qwen3-4B-it2026.06 | 93 | |
| No SteeringModel=Qwen3-4B-it2026.06 | 76 | |
| Target featureModel=Qwen3-1.7B-it2026.06 | 66.7 | |
| No SteeringModel=Qwen3-1.7B-it2026.06 | 41.7 | |
| Random featureModel=Qwen3-4B-it2026.06 | 1.2 | |
| Random featureModel=Qwen3-1.7B-it2026.06 | 0.3 |