Jailbreak attacks on MML-Mirror (MML-M)
97Safety RateTRR
Evaluation Results
| Method | Links | |
|---|---|---|
| TRRBackbone=Qwen2.5VL-7B, Alignment Method=TRR2025.12 | 97 | |
| TRRBackbone=Qwen2.5VL-32B, Alignment Method=TRR2025.12 | 80.5 | |
| TiSBackbone=Qwen2.5VL-32B, Alignment Method=TiS2025.12 | 65 | |
| MSR-AlignBackbone=Qwen2.5VL-32B, Alignment Method=MSR-Align2025.12 | 55.7 | |
| MSR-AlignBackbone=Qwen2.5VL-7B, Alignment Method=MSR-Align2025.12 | 53.7 | |
| Claude-3.5-SonnetModel=Claude-3.5-Sonnet2025.12 | 40 | |
| TiSBackbone=Qwen2.5VL-7B, Alignment Method=TiS2025.12 | 28.8 | |
| Gemini-2.5-proModel=Gemini-2.5-pro2025.12 | 15.7 | |
| Safework-R1-7BModel=Safework-R1-7B2025.12 | 12.5 | |
| Qwen2.5VL-7BBackbone=Qwen2.5VL-7B, Alignment Method=None2025.12 | 6.5 | |
| GPT-4oModel=GPT-4o2025.12 | 1.9 | |
| Qwen2.5VL-32BBackbone=Qwen2.5VL-32B, Alignment Method=None2025.12 | 0.7 |