Red-teaming on AdversarialQA
0ASRSafeTransformer
Evaluation results
| Method | Links | |
|---|---|---|
| SafeTransformerPrompt=CoU, Inference Mode=eos2026.03 | 0 | |
| SafeTransformerPrompt=CoU, Inference Mode=avg2026.03 | 0 | |
| SafeTransformerPrompt=CoT, Inference Mode=eos2026.03 | 2.4 | |
| SafeTransformerPrompt=Standard, Inference Mode=eos2026.03 | 2.8 | |
| SafeTransformerPrompt=Standard, Inference Mode=avg2026.03 | 3.4 | |
| SafeTransformerPrompt=CoT, Inference Mode=avg2026.03 | 3.8 | |
| Llama-3.2-1B (SFT)Prompt=Suffix2026.03 | 5.8 | |
| Llama-3.2-1B (SFT)Prompt=Standard2026.03 | 6.4 | |
| SafeTransformerPrompt=Avg., Inference Mode=eos2026.03 | 6.4 | |
| SafeTransformerPrompt=Avg., Inference Mode=avg2026.03 | 6.45 | |
| Llama-3.2-1B (SFT)Prompt=CoU2026.03 | 6.6 | |
| Llama-3.2-1B (SFT)Prompt=Avg.2026.03 | 10.05 | |
| Llama-3.2-1BPrompt=CoU2026.03 | 10.8 | |
| SafeTransformerPrompt=Suffix, Inference Mode=avg2026.03 | 18.6 | |
| SafeTransformerPrompt=Suffix, Inference Mode=eos2026.03 | 20.4 | |
| Llama-3.2-1B (SFT)Prompt=CoT2026.03 | 21.4 | |
| Llama-3.2-1BPrompt=Avg.2026.03 | 27.3 | |
| Llama-3.2-1BPrompt=Suffix2026.03 | 29.6 | |
| Llama-3.2-1BPrompt=Standard2026.03 | 33.2 | |
| Llama-3.2-1BPrompt=CoT2026.03 | 35.6 |