Benign Request Processing on Aggregated Benign Request Datasets
0.1Benign Macro FPRNo defense
Evaluation Results
| Method | Links | |
|---|---|---|
| No defenseBase Model=Qwen2.5-72B2026.06 | 0.1 | |
| ICDBase Model=Qwen2.5-72B2026.06 | 0.5 | |
| Self Reminder-OPBase Model=Qwen2.5-72B2026.06 | 1.1 | |
| RAINBase Model=Qwen2.5-72B2026.06 | 1.1 | |
| DoubtProbeBase Model=Qwen2.5-72B2026.06 | 1.9 | |
| Self Reminder-HPBase Model=Qwen2.5-72B2026.06 | 2.8 | |
| SelfDefend-IntentBase Model=Qwen2.5-72B2026.06 | 3.5 | |
| SelfDefend-DirectBase Model=Qwen2.5-72B2026.06 | 4.3 | |
| LlamaGuard4Base Model=Qwen2.5-72B2026.06 | 5.3 | |
| Goal PrioritizationBase Model=Qwen2.5-72B2026.06 | 5.4 | |
| LlamaGuard2Base Model=Qwen2.5-72B2026.06 | 59.1 | |
| LlamaGuard3Base Model=Qwen2.5-72B2026.06 | 80.9 | |
| SmoothLLMBase Model=Qwen2.5-72B2026.06 | 96.4 |