Safety Alignment Evaluation on VLSBench
100Safety ScoreGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOBackbone=Qwen3-VL-4B, Alignment Method=GRPO2026.06 | 100 | |
| COPSD (Hybrid)Backbone=Qwen3-VL-4B, Alignment Method=COPSD (Hybrid)2026.06 | 99.51 | |
| Safe-RLHF-VBackbone=Qwen3-VL-4B, Alignment Method=Safe-RLHF-V2026.06 | 99.33 | |
| GRPOBackbone=Qwen2.5-VL-7B, Alignment Method=GRPO2026.06 | 98.93 | |
| COPSD (Standard)Backbone=Qwen3-VL-4B, Alignment Method=COPSD (Standard)2026.06 | 98.75 | |
| Safe-RLHF-VBackbone=Qwen2.5-VL-7B, Alignment Method=Safe-RLHF-V2026.06 | 98.53 | |
| COPSD (Standard)Backbone=Qwen2.5-VL-7B, Alignment Method=COPSD (Standard)2026.06 | 97.72 | |
| COPSD (Hybrid)Backbone=Qwen2.5-VL-7B, Alignment Method=COPSD (Hybrid)2026.06 | 94.82 | |
| OPDBackbone=Qwen2.5-VL-7B, Alignment Method=OPD2026.06 | 90.36 | |
| SFTBackbone=Qwen2.5-VL-7B, Alignment Method=SFT2026.06 | 90.14 | |
| SFTBackbone=Qwen3-VL-4B, Alignment Method=SFT2026.06 | 79.92 | |
| OPDBackbone=Qwen3-VL-4B, Alignment Method=OPD2026.06 | 79.61 | |
| Qwen3-VL-4BBackbone=Qwen3-VL-4B, Alignment Method=Base2026.06 | 68.63 | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B, Alignment Method=Base2026.06 | 62.92 |