Multi-turn MLLM Safety Evaluation on STEER-SPA Multi-turn
66.67SafetySAFER-STEER
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SAFER-STEERBackbone=Qwen2.5VL-7B, Alignment Strategy=SAFER-STEER2026.03 | 66.67 | 70.93 | |
| SaFeR-VLMBackbone=Qwen2.5VL-7B, Alignment Strategy=SaFeR-VLM2026.03 | 61.13 | 61.13 | |
| SAFER-STEERBackbone=Qwen2.5VL-3B, Alignment Strategy=SAFER-STEER2026.03 | 54.76 | 66.27 | |
| SFTBackbone=Qwen2.5VL-7B, Alignment Strategy=SFT2026.03 | 51.32 | 65.66 | |
| SFTBackbone=Qwen2.5VL-3B, Alignment Strategy=SFT2026.03 | 48.68 | 58.11 | |
| SaFeR-VLMBackbone=Qwen2.5VL-3B, Alignment Strategy=SaFeR-VLM2026.03 | 41.13 | 52.08 | |
| SPA-VLBackbone=Qwen2.5VL-7B, Alignment Strategy=SPA-VL2026.03 | 39.85 | 52.87 | |
| TISBackbone=Qwen2.5VL-7B, Alignment Strategy=TIS2026.03 | 32.58 | 22.35 | |
| Qwen2.5VL-7BBackbone=Qwen2.5VL-7B, Alignment Strategy=Base2026.03 | 29.01 | 46.56 | |
| TISBackbone=Qwen2.5VL-3B, Alignment Strategy=TIS2026.03 | 28.24 | 22.14 | |
| SPA-VLBackbone=Qwen2.5VL-3B, Alignment Strategy=SPA-VL2026.03 | 25.86 | 33.84 | |
| MMDPOBackbone=Qwen2.5VL-7B, Alignment Strategy=MMDPO2026.03 | 25 | 50.42 | |
| GRLHF-VBackbone=Qwen2.5VL-7B, Alignment Strategy=GRLHF-V2026.03 | 17.11 | 35.36 | |
| SafeRLHF-VBackbone=Qwen2.5VL-7B, Alignment Strategy=SafeRLHF-V2026.03 | 16.29 | 18.18 | |
| Qwen2.5VL-3BBackbone=Qwen2.5VL-3B, Alignment Strategy=Base2026.03 | 14.89 | 23.28 | |
| MMDPOBackbone=Qwen2.5VL-3B, Alignment Strategy=MMDPO2026.03 | 10.57 | 21.13 | |
| GRLHF-VBackbone=Qwen2.5VL-3B, Alignment Strategy=GRLHF-V2026.03 | 8.71 | 23.48 | |
| SafeRLHF-VBackbone=Qwen2.5VL-3B, Alignment Strategy=SafeRLHF-V2026.03 | 6.06 | 10.98 |