Safety Classification on XSTest, OpenAI, ToxicC., BeaverT., RTP-LX, XSafety German
82.9XSTest AccuracyLlamaGuard 3
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LlamaGuard 3Size=8B2025.02 | 82.9 | 78.5 | 48 | 70.4 | 50.2 | 37.8 | 61.3 | |
| LlamaGuard 2Size=8B2025.02 | 79.8 | 74.4 | 40.5 | 68.5 | 38.7 | 30.6 | 55.4 | |
| DuoGuardSize=0.5B2025.02 | 75.8 | 65.9 | 61.4 | 80.8 | 87.3 | 60.4 | 71.9 | |
| ShieldGemmaSize=2B2025.02 | 59.6 | 38.7 | 27.5 | 51.6 | 19.5 | 24.1 | 36.8 | |
| LlamaGuard 3Size=1B2025.02 | 43 | 37.4 | 20.9 | 50.2 | 55.4 | 61.4 | 44.7 |