Harmful Refusal on HarmBench Standard
98.2Refusal Rate (String Match)LlamaGuard prompt-filter
Evaluation Results
| Method | Links | |
|---|---|---|
| LlamaGuard prompt-filterDraft Model=Llama-3.1-8B2026.06 | 98.2 | |
| LlamaGuard response-filterDraft Model=Llama-3.1-8B2026.06 | 98.2 | |
| Hard prefixDraft Model=Llama-3.1-8B, K (Beam Count)=12026.06 | 82 | |
| ALIGNBEAMDraft Model=Llama-3.1-8B, K (Beam Count)=32026.06 | 79.6 | |
| Base draftDraft Model=Llama-3.1-8B2026.06 | 15 | |
| Safety promptDraft Model=Llama-3.1-8B2026.06 | 10.2 |