Policy-adaptive safety guardrailing on POLICYSHIFTBENCH (Shift)
67F1 ScorePolicyShiftGuard-7B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PolicyShiftGuard-7BParameter Scale=7B2026.07 | 67 | 69.9 | |
| GuardReasoner-VL-3BParameter Scale=3B2026.07 | 54.7 | 51.8 | |
| GuardReasoner-VL-7BParameter Scale=7B2026.07 | 54.2 | 48.2 | |
| PolicyShiftGuard-3BParameter Scale=3B2026.07 | 47.8 | 62.4 | |
| ShieldGemma2-4BParameter Scale=4B2026.07 | 45.4 | 36.8 | |
| QwenGuard-7BParameter Scale=7B2026.07 | 39.5 | 38.7 | |
| SafeGuard-VL-RL-7BParameter Scale=7B2026.07 | 39.5 | 52.5 | |
| Qwen2.5-VL-7BParameter Scale=7B2026.07 | 12.1 | 32.9 | |
| LlamaGuard4-12BParameter Scale=12B2026.07 | 9.9 | 19.1 |