Safety Classification on XSTest, OpenAI, ToxicC., BeaverT., RTP-LX, XSafety Suite
88.8XSTest AccuracyLlamaGuard 2
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LlamaGuard 2Size=8B2025.02 | 88.8 | 75.9 | 46.3 | 72.3 | 39.5 | 35.2 | 59.7 | |
| LlamaGuard 3Size=8B2025.02 | 88.4 | 79 | 54 | 70.1 | 48.5 | 40.5 | 63.4 | |
| DuoGuardSize=0.5B2025.02 | 82.3 | 70.8 | 70.1 | 86.1 | 91.7 | 48.5 | 74.9 | |
| ShieldGemmaSize=2B2025.02 | 69.4 | 44.8 | 36.4 | 51.6 | 26 | 30.6 | 43.1 | |
| LlamaGuard 3Size=1B2025.02 | 43.4 | 36.8 | 22.3 | 51.6 | 54.6 | 62.3 | 45.2 |