Benign Over-refusal on JBB (Benign)
10String-match Refusal Rate (%)Base draft
Evaluation Results
| Method | Links | |
|---|---|---|
| Base draftDraft Model=Llama-3.1-8B2026.06 | 10 | |
| Safety promptDraft Model=Llama-3.1-8B2026.06 | 10 | |
| ALIGNBEAMDraft Model=Llama-3.1-8B, K (Beam Count)=32026.06 | 26 | |
| Hard prefixDraft Model=Llama-3.1-8B, K (Beam Count)=12026.06 | 29 | |
| LlamaGuard prompt-filterDraft Model=Llama-3.1-8B2026.06 | 38 | |
| LlamaGuard response-filterDraft Model=Llama-3.1-8B2026.06 | 39 |