Safety Classification on XSTest (test)
92.91F1LEG large
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LEG largeTrain Dataset=WildGuardMix, Model Size=304M2026.01 | 92.91 | — | — | — | — | |
| OracleCondition=Perfect Routing2025.02 | 91.9 | 97.1 | 87.2 | 16.68 | — | |
| SafeRouteRouting=SafeRoute (Ours), Small Model=Llama-Guard-3-1B, Large Model=Granite-Guardian-3-8B2025.02 | 89.2 | 97 | 82.6 | 4.28 | — | |
| EntRouting=Entropy-based, Small Model=Llama-Guard-3-1B, Large Model=Granite-Guardian-3-8B2025.02 | 88.9 | 97 | 82.1 | 6.12 | — | |
| CCRouting=Calibration-based, Small Model=Llama-Guard-3-1B, Large Model=Granite-Guardian-3-8B2025.02 | 88.9 | 97 | 82.1 | 5.94 | — | |
| Llama Guard 2Train Dataset=N/A, Model Size=8B2026.01 | 88.88 | — | — | — | — | |
| Llama Guard 3Train Dataset=N/A, Model Size=8B2026.01 | 88.4 | — | — | — | — | |
| BCRouting=Bayesian-based, Small Model=Llama-Guard-3-1B, Large Model=Granite-Guardian-3-8B2025.02 | 87.4 | 90.4 | 84.6 | 6.47 | — | |
| RandomRouting=Random, Small Model=Llama-Guard-3-1B, Large Model=Granite-Guardian-3-8B2025.02 | 86.1 | 89.3 | 81.3 | 9.36 | — | |
| Granite-Guardian-3-8BModel=Large (Granite-Guardian-3-8B)2025.02 | 85.7 | 91.3 | 80.8 | 12.85 | — | |
| LEG baseTrain Dataset=WildGuardMix, Model Size=86M2026.01 | 85.56 | — | — | — | — | |
| TSRouting=Thresholding, Small Model=Llama-Guard-3-1B, Large Model=Granite-Guardian-3-8B2025.02 | 85.5 | 92.5 | 79.5 | 12.9 | — | |
| Llama-Guard-3-1BModel=Small (Llama-Guard-3-1B)2025.02 | 85.3 | 88.9 | 82.1 | 3.47 | — | |
| DuoGuardTrain Dataset=N/A, Model Size=0.5B2026.01 | 82.3 | — | — | — | — | |
| LEG largeTrain Dataset=AEGIS2.0, Model Size=304M2026.01 | 77.99 | — | — | — | — | |
| LEG largeTrain Dataset=ToxicChat0124, Model Size=304M2026.01 | 77.09 | — | — | — | — | |
| LEG baseTrain Dataset=AEGIS2.0, Model Size=86M2026.01 | 75 | — | — | — | — | |
| LEG baseTrain Dataset=ToxicChat0124, Model Size=86M2026.01 | 72.73 | — | — | — | — | |
| ShieldGammaTrain Dataset=N/A, Model Size=2B2026.01 | 69.4 | — | — | — | — | |
| Llama Guard 3Train Dataset=N/A, Model Size=1B2026.01 | 43.4 | — | — | — | — | |
| Latent GuardBase Model=LLAMA2-CHAT-7B2025.07 | — | — | — | — | 100 | |
| Latent GuardBase Model=LLAMA3-INSTRUCT-8B2025.07 | — | — | — | — | 78.5 | |
| Latent GuardBase Model=QWEN2-INSTRUCT-7B2025.07 | — | — | — | — | 91.6 | |
| Llama Guard 3Base Model=LLAMA2-CHAT-7B2025.07 | — | — | — | — | 84.4 | |
| Llama Guard 3Base Model=LLAMA3-INSTRUCT-8B2025.07 | — | — | — | — | 50 | |
| Llama Guard 3Base Model=QWEN2-INSTRUCT-7B2025.07 | — | — | — | — | 50 |