Safety and Helpfulness Evaluation on Beavertails
92.03Safety ScoreSAFER-STEER
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAFER-STEERBase Model=Qwen2.5VL-7B, Method Category=Dynamic Methods2026.03 | 92.03 | 86.83 | |
| SAFER-STEERBase Model=Qwen2.5VL-3B, Method Category=Dynamic Methods2026.03 | 88.89 | 85.97 | |
| SFTBase Model=Qwen2.5VL-7B, Method Category=Dynamic Methods2026.03 | 88.81 | 86.1 | |
| SFTBase Model=Qwen2.5VL-3B, Method Category=Dynamic Methods2026.03 | 86.95 | 82.88 | |
| TISBase Model=Qwen2.5VL-7B, Method Category=Static Methods2026.03 | 85.59 | 77.8 | |
| SPA-VLBase Model=Qwen2.5VL-7B, Method Category=Static Methods2026.03 | 80.85 | 77.76 | |
| SPA-VLBase Model=Qwen2.5VL-3B, Method Category=Static Methods2026.03 | 80.81 | 76.36 | |
| TISBase Model=Qwen2.5VL-3B, Method Category=Static Methods2026.03 | 79.15 | 73.73 | |
| SaFeR-VLMBase Model=Qwen2.5VL-7B, Method Category=Static Methods2026.03 | 76.78 | 75.93 | |
| SaFeR-VLMBase Model=Qwen2.5VL-3B, Method Category=Static Methods2026.03 | 76.44 | 81.19 | |
| MMDPOBase Model=Qwen2.5VL-7B, Method Category=Static Methods2026.03 | 68.54 | 73.03 | |
| Qwen2.5VL-7BBase Model=Qwen2.5VL-7B, Method Category=Base2026.03 | 55.93 | 58.98 | |
| GRLHF-VBase Model=Qwen2.5VL-7B, Method Category=Static Methods2026.03 | 49.49 | 55.59 | |
| MMDPOBase Model=Qwen2.5VL-3B, Method Category=Static Methods2026.03 | 47.18 | 54.19 | |
| SafeRLHF-VBase Model=Qwen2.5VL-7B, Method Category=Static Methods2026.03 | 46.1 | 38.81 | |
| Qwen2.5VL-3BBase Model=Qwen2.5VL-3B, Method Category=Base2026.03 | 35.31 | 41.93 | |
| GRLHF-VBase Model=Qwen2.5VL-3B, Method Category=Static Methods2026.03 | 32.54 | 36.95 | |
| SafeRLHF-VBase Model=Qwen2.5VL-3B, Method Category=Static Methods2026.03 | 29.15 | 24.07 |