Prompt-Response Safety Routing on HarmBench
55.92Routing F1SafeRoute
Evaluation Results
| Method | Links | |
|---|---|---|
| SafeRoutesmall_model=Llama-Guard-3-1B, large_model=Granite-Guardian-3-8B2025.02 | 55.92 | |
| SafeRouteProtocol=Prompt-Response, Small Model=Llama-Guard-3-1B, Large Model=Llama-Guard-3-8B2025.02 | 51.24 | |
| Entropysmall_model=Llama-Guard-3-1B, large_model=Granite-Guardian-3-8B2025.02 | 43.45 | |
| EntropyProtocol=Prompt-Response, Small Model=Llama-Guard-3-1B, Large Model=Llama-Guard-3-8B2025.02 | 40.94 | |
| +CCsmall_model=Llama-Guard-3-1B, large_model=Granite-Guardian-3-8B2025.02 | 38.28 | |
| CCProtocol=Prompt-Response, Small Model=Llama-Guard-3-1B, Large Model=Llama-Guard-3-8B2025.02 | 37.86 | |
| BCProtocol=Prompt-Response, Small Model=Llama-Guard-3-1B, Large Model=Llama-Guard-3-8B2025.02 | 32.62 | |
| +BCsmall_model=Llama-Guard-3-1B, large_model=Granite-Guardian-3-8B2025.02 | 28.46 | |
| +TSsmall_model=Llama-Guard-3-1B, large_model=Granite-Guardian-3-8B2025.02 | 22.64 | |
| TSProtocol=Prompt-Response, Small Model=Llama-Guard-3-1B, Large Model=Llama-Guard-3-8B2025.02 | 19.3 |