Safety Classification on XSTest, OpenAI, ToxicC., BeaverT., RTP-LX, XSafety Spanish
86.2XSTest AccuracyLlamaGuard 3
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LlamaGuard 3Size=8B2025.02 | 86.2 | 77.7 | 48.4 | 69.5 | 48.4 | 39 | 61.5 | |
| LlamaGuard 2Size=8B2025.02 | 84 | 74.8 | 39.2 | 67.5 | 39.4 | 33.8 | 56.5 | |
| DuoGuardSize=0.5B2025.02 | 81.4 | 66.8 | 64.9 | 81.4 | 88 | 61 | 73.9 | |
| ShieldGemmaSize=2B2025.02 | 62.4 | 37.7 | 29.1 | 50.8 | 17.8 | 24 | 37 | |
| LlamaGuard 3Size=1B2025.02 | 46.9 | 37.9 | 20.4 | 50.3 | 52.1 | 62.1 | 45 |