Safety Evaluation on SIUO (test)
95.67Effective ScoreSAFE RLHF-v
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAFE RLHF-vBackbone=Qwen2.5-VL-7B2026.02 | 95.67 | 55.9 | |
| Pragma-VLBackbone=Qwen2.5-VL-7B2026.02 | 95.21 | 63.47 | |
| Beavertails-V helpBackbone=Qwen2.5-VL-7B2026.02 | 95.2 | 34.33 | |
| Pragma-VLBackbone=Llava-1.5-7B2026.02 | 94.01 | 55.42 | |
| SPA-VLBackbone=Qwen2.5-VL-7B2026.02 | 93.37 | 36.74 | |
| SFTBackbone=Qwen2.5-VL-7B2026.02 | 93.29 | 49.39 | |
| Beavertails-V allBackbone=Qwen2.5-VL-7B2026.02 | 92.59 | 51.23 | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.02 | 92.17 | 38.78 | |
| DPOBackbone=Qwen2.5-VL-7B2026.02 | 91.61 | 59.03 | |
| Llava-1.5-7BBackbone=Llava-1.5-7B2026.02 | 90.41 | 14.37 | |
| Beavertails-V harmBackbone=Qwen2.5-VL-7B2026.02 | 89.76 | 59.64 | |
| Beavertails-V helpBackbone=Llava-1.5-7B2026.02 | 88.62 | 30.53 | |
| SPA-VLBackbone=Llava-1.5-7B2026.02 | 86.83 | 36.53 | |
| SAFE RLHF-vBackbone=Llava-1.5-7B2026.02 | 86.41 | 47.53 | |
| SFTBackbone=Llava-1.5-7B2026.02 | 86.22 | 47.3 | |
| Beavertails-V allBackbone=Llava-1.5-7B2026.02 | 82.72 | 41.35 | |
| MM-RLHFBackbone=Qwen2.5-VL-7B2026.02 | 80.23 | 51.8 | |
| DPOBackbone=Llava-1.5-7B2026.02 | 78.31 | 44.91 | |
| MM-RLHFBackbone=Llava-1.5-7B2026.02 | 73.65 | 37.95 | |
| Beavertails-V harmBackbone=Llava-1.5-7B2026.02 | 70.05 | 32.28 |