Multimodal Content Moderation on UnsafeBench Sexual Text-Only
81.82AccuracyKidsNanny
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| KidsNannyRegime=2, Stage=1+22026.03 | 81.82 | 86.21 | 75.76 | 100 | 122 | |
| ShieldGemma-2Regime=22026.03 | 59.09 | 70 | 60 | 84 | 1,136 | |
| LlavaGuardRegime=22026.03 | 59.09 | 60.87 | 66.67 | 56 | 4,138 |