Safety Evaluation on SafeMT
82.1HelpfulnessAM3Safety
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AM3SafetyBackbone=LLaVA-NeXT-7B2026.01 | 82.1 | 69.18 | 38.44 | |
| MM-DPOBackbone=Qwen2.5-VL-7B2026.01 | 63.78 | 51.79 | 44.75 | |
| AM3SafetyBackbone=Qwen2.5-VL-7B2026.01 | 63.19 | 58.19 | 28.06 | |
| RLHF-VBackbone=LLaVA-NeXT-7B2026.01 | 59.41 | 50.35 | 55.7 | |
| Safe RLHF-VBackbone=Qwen2.5-VL-7B2026.01 | 57.18 | 52.17 | 42.15 | |
| MM-DPOBackbone=LLaVA-NeXT-7B2026.01 | 56.13 | 54.36 | 47.05 | |
| Qwen2.5-VL-7BAlignment=Original Base Model2026.01 | 50 | 50 | 48.92 | |
| LLaVA-NeXT-7BAlignment=Original Base Model2026.01 | 50 | 50 | 48.95 | |
| SPA-VLBackbone=LLaVA-NeXT-7B2026.01 | 48.36 | 48.66 | 32.25 | |
| Safe RLHF-VBackbone=LLaVA-NeXT-7B2026.01 | 47.44 | 48.91 | 49.1 | |
| SPA-VLBackbone=Qwen2.5-VL-7B2026.01 | 33.47 | 45.15 | 30.9 | |
| RLHF-VBackbone=Qwen2.5-VL-7B2026.01 | 18.55 | 30.56 | 42.55 |