Multimodal Safety Evaluation on SPA-VL
91.89Safety ScoreSaFeR-ToolKit (+ SFT+DPO+GRPO) [3B]
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SaFeR-ToolKit (+ SFT+DPO+GRPO) [3B]Model Scale=3B, Training Configuration=SFT+DPO+GRPO2026.03 | 91.89 | 85.09 | 87.36 | |
| SaFeR-ToolKit (+ SFT+DPO+GRPO) [7B]Model Scale=7B, Training Configuration=SFT+DPO+GRPO2026.03 | 91.89 | 89.25 | 90.94 | |
| Qwen2.5-VL-7BModel Scale=7B2026.03 | 76.42 | 73.21 | 40 | |
| Qwen2.5-VL-3BModel Scale=3B2026.03 | 46.6 | 54.53 | 11.89 | |
| Qwen2.5-VL-3B + SFT+DPO+GRPOBase Model=Qwen2.5-VL-3B, Stage=SFT+DPO+GRPO2026.03 | 2.81 | 2.8 | 2.84 | |
| Qwen2.5-VL-7B + TISBase Model=Qwen2.5-VL-7B, Enhancement=TIS2026.03 | 2.8 | 2.68 | 2.72 | |
| Qwen2.5-VL-7B + SPA-VLBase Model=Qwen2.5-VL-7B, Enhancement=SPA-VL2026.03 | 2.8 | 2.7 | 2.28 | |
| Qwen2.5-VL-3B + SFT+GRPOBase Model=Qwen2.5-VL-3B, Stage=SFT+GRPO2026.03 | 2.78 | 2.7 | 2.82 | |
| Qwen2.5-VL-7B + SFT+DPO+GRPOBase Model=Qwen2.5-VL-7B, Stage=SFT+DPO+GRPO2026.03 | 2.77 | 2.87 | 2.88 | |
| Qwen2.5-VL-7B + SFT+GRPOBase Model=Qwen2.5-VL-7B, Stage=SFT+GRPO2026.03 | 2.75 | 2.84 | 2.88 | |
| Qwen2.5-VL-7B + SaFeR-VLMBase Model=Qwen2.5-VL-7B, Method=SaFeR-VLM2026.03 | 2.73 | 2.4 | 2.46 | |
| Qwen2.5-VL-3B + SPA-VLBase Model=Qwen2.5-VL-3B, Enhancement=SPA-VL2026.03 | 2.72 | 2.65 | 2.2 | |
| Qwen2.5-VL-7B + SFT+DPOBase Model=Qwen2.5-VL-7B, Stage=SFT+DPO2026.03 | 2.72 | 2.72 | 2.86 | |
| Qwen2.5-VL-7B + SFTBase Model=Qwen2.5-VL-7B, Stage=SFT2026.03 | 2.68 | 2.65 | 2.78 | |
| Qwen2.5-VL-3B + TISBase Model=Qwen2.5-VL-3B, Enhancement=TIS2026.03 | 2.62 | 2.35 | 2.5 | |
| Qwen2.5-VL-3B + SaFeR-VLMBase Model=Qwen2.5-VL-3B, Method=SaFeR-VLM2026.03 | 2.62 | 2.31 | 2.27 | |
| Qwen2.5-VL-3B + SFT+DPOBase Model=Qwen2.5-VL-3B, Stage=SFT+DPO2026.03 | 2.6 | 2.54 | 2.67 | |
| Qwen2.5-VL-3B + SFTBase Model=Qwen2.5-VL-3B, Stage=SFT2026.03 | 2.52 | 2.46 | 2.63 | |
| Qwen2.5-VL-7B + VLGuardBase Model=Qwen2.5-VL-7B, Enhancement=VLGuard2026.03 | 2.49 | 1.67 | 1.8 | |
| Qwen2.5-VL-7B + SIABase Model=Qwen2.5-VL-7B, Enhancement=SIA2026.03 | 2.46 | 2.01 | 1.94 | |
| Qwen2.5-VL-3B + VLGuardBase Model=Qwen2.5-VL-3B, Enhancement=VLGuard2026.03 | 2.43 | 1.63 | 1.74 | |
| Qwen2.5-VL-7B + ECSOBase Model=Qwen2.5-VL-7B, Enhancement=ECSO2026.03 | 2.37 | 2.64 | 2.27 | |
| Qwen2.5-VL-7BBase Model=Qwen2.5-VL-7B2026.03 | 2.27 | 2.66 | 2.27 | |
| Qwen2.5-VL-3B + SIABase Model=Qwen2.5-VL-3B, Enhancement=SIA2026.03 | 1.7 | 1.63 | 1.38 | |
| Qwen2.5-VL-3B + ECSOBase Model=Qwen2.5-VL-3B, Enhancement=ECSO2026.03 | 1.39 | 2.16 | 1.66 | |
| Qwen2.5-VL-3BBase Model=Qwen2.5-VL-3B2026.03 | 1.38 | 2.35 | 1.73 |