Safety Evaluation on MMSafe-PO
69.64HelpfulnessAM3Safety
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AM3SafetyBackbone=LLaVA-NeXT-7B2026.01 | 69.64 | 65.62 | 10.55 | |
| AM3SafetyBackbone=Qwen2.5-VL-7B2026.01 | 57.02 | 54.82 | 12 | |
| MM-DPOBackbone=LLaVA-NeXT-7B2026.01 | 52.36 | 56.34 | 15.64 | |
| MM-DPOBackbone=Qwen2.5-VL-7B2026.01 | 52.17 | 52.64 | 14.91 | |
| Qwen2.5-VL-7BAlignment=Original Base Model2026.01 | 50 | 50 | 14.55 | |
| LLaVA-NeXT-7BAlignment=Original Base Model2026.01 | 50 | 50 | 17.82 | |
| SPA-VLBackbone=LLaVA-NeXT-7B2026.01 | 48.05 | 61.4 | 14.55 | |
| Safe RLHF-VBackbone=Qwen2.5-VL-7B2026.01 | 47.38 | 51.97 | 15.64 | |
| Safe RLHF-VBackbone=LLaVA-NeXT-7B2026.01 | 45.37 | 46.25 | 16.36 | |
| SPA-VLBackbone=Qwen2.5-VL-7B2026.01 | 44.47 | 51.09 | 14.55 | |
| RLHF-VBackbone=LLaVA-NeXT-7B2026.01 | 13.6 | 15.71 | 24 | |
| RLHF-VBackbone=Qwen2.5-VL-7B2026.01 | 10.42 | 13.91 | 24.73 |