Multi-turn MLLM Safety Evaluation on STEER-DyS Multi-turn
61.36Safety ScoreSAFER-STEER
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAFER-STEERBackbone=Qwen2.5VL-7B, Alignment Strategy=SAFER-STEER2026.03 | 61.36 | 55.42 | |
| SFTBackbone=Qwen2.5VL-7B, Alignment Strategy=SFT2026.03 | 50.46 | 51.68 | |
| SAFER-STEERBackbone=Qwen2.5VL-3B, Alignment Strategy=SAFER-STEER2026.03 | 49.4 | 55.4 | |
| SaFeR-VLMBackbone=Qwen2.5VL-7B, Alignment Strategy=SaFeR-VLM2026.03 | 49.38 | 47.85 | |
| SFTBackbone=Qwen2.5VL-3B, Alignment Strategy=SFT2026.03 | 35.17 | 38.84 | |
| SaFeR-VLMBackbone=Qwen2.5VL-3B, Alignment Strategy=SaFeR-VLM2026.03 | 32.42 | 35.17 | |
| TISBackbone=Qwen2.5VL-7B, Alignment Strategy=TIS2026.03 | 32.16 | 21.29 | |
| SPA-VLBackbone=Qwen2.5VL-7B, Alignment Strategy=SPA-VL2026.03 | 29.47 | 35.58 | |
| MMDPOBackbone=Qwen2.5VL-7B, Alignment Strategy=MMDPO2026.03 | 26.64 | 44.47 | |
| TISBackbone=Qwen2.5VL-3B, Alignment Strategy=TIS2026.03 | 24.92 | 16.82 | |
| Qwen2.5VL-7BBackbone=Qwen2.5VL-7B, Alignment Strategy=Base2026.03 | 22.27 | 38.7 | |
| GRLHF-VBackbone=Qwen2.5VL-7B, Alignment Strategy=GRLHF-V2026.03 | 18.87 | 34.66 | |
| SPA-VLBackbone=Qwen2.5VL-3B, Alignment Strategy=SPA-VL2026.03 | 16.74 | 19.69 | |
| SafeRLHF-VBackbone=Qwen2.5VL-7B, Alignment Strategy=SafeRLHF-V2026.03 | 13.43 | 12.5 | |
| MMDPOBackbone=Qwen2.5VL-3B, Alignment Strategy=MMDPO2026.03 | 11.33 | 18.22 | |
| GRLHF-VBackbone=Qwen2.5VL-3B, Alignment Strategy=GRLHF-V2026.03 | 9.63 | 20.64 | |
| Qwen2.5VL-3BBackbone=Qwen2.5VL-3B, Alignment Strategy=Base2026.03 | 9.4 | 19.44 | |
| SafeRLHF-VBackbone=Qwen2.5VL-3B, Alignment Strategy=SafeRLHF-V2026.03 | 5.25 | 6.33 |