Harmfulness Refusal on HarmBench Risk 3: Harmful Reduction
0Attack Success RateM+
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| M+Base Model=Gemma-9B2025.11 | 0 | 9.0158 | |
| M'Base Model=Gemma-9B2025.11 | 0 | 9.0158 | |
| M+Base Model=Mistral-7B2025.11 | 0 | 9.2847 | |
| M'Base Model=Mistral-7B2025.11 | 0 | 9.2847 | |
| M+Base Model=Llama-3-8B2025.11 | 0 | 8.7634 | |
| M'Base Model=Llama-3-8B2025.11 | 0 | 8.7634 | |
| M_safepromptBase Model=Llama-3-8B2025.11 | 28.1 | 8.4592 | |
| M_safepromptBase Model=Gemma-9B2025.11 | 40.3 | 8.6734 | |
| M_safepromptBase Model=Mistral-7B2025.11 | 57.4 | 8.9421 | |
| MBase Model=Gemma-9B2025.11 | 68 | 2.2545 | |
| MBase Model=Llama-3-8B2025.11 | 68 | 2.3179 | |
| MBase Model=Mistral-7B2025.11 | 70 | 2.1823 |