Targeted Attack Detection on Alpaca AddTarget Medium
100TPRConfGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ConfGuardModel=Qwen2.5 7B, Attacks=Phrase2026.03 | 100 | 27.01 | |
| ConfGuardModel=Qwen2.5 7B, Attacks=Word2026.03 | 99.88 | 26.16 | |
| ConfGuardModel=Qwen2.5 14B, Attacks=System2026.03 | 99.6 | 15 | |
| PPLModel=Qwen2.5 7B, Attacks=Joint2026.03 | 99.58 | 34.67 | |
| ConfGuardModel=Qwen2.5 7B, Attacks=Joint2026.03 | 99.58 | 27.41 | |
| DualSentinelModel=Qwen2.5 7B, Attacks=Joint2026.03 | 99.58 | 12.95 | |
| PPLModel=Qwen2.5 7B, Attacks=Phrase2026.03 | 99.54 | 34.67 | |
| DualSentinelModel=Qwen2.5 7B, Attacks=Phrase2026.03 | 99.54 | 8.5 | |
| PPLModel=Qwen2.5 14B, Attacks=System2026.03 | 99.5 | 30.08 | |
| DualSentinelModel=Qwen2.5 7B, Attacks=Word2026.03 | 99.21 | 10.23 | |
| DualSentinelModel=Qwen2.5 14B, Attacks=System2026.03 | 99.2 | 3 | |
| ConfGuardModel=Qwen2.5 7B, Attacks=Sentence2026.03 | 98.66 | 28.34 | |
| STRIPModel=Qwen2.5 7B, Attacks=Joint2026.03 | 98.32 | 88.3 | |
| STRIPModel=Qwen2.5 7B, Attacks=Word2026.03 | 97.75 | 85.82 | |
| STRIPModel=Qwen2.5 7B, Attacks=Sentence2026.03 | 97.33 | 88.25 | |
| CleanGenModel=Qwen2.5 7B, Attacks=Sentence2026.03 | 97.33 | 78.13 | |
| DualSentinelModel=Qwen2.5 7B, Attacks=Sentence2026.03 | 97.33 | 12.6 | |
| STRIPModel=Qwen2.5 7B, Attacks=Phrase2026.03 | 97.26 | 85.88 | |
| CleanGenModel=Qwen2.5 7B, Attacks=Word2026.03 | 94.17 | 76.92 | |
| CleanGenModel=Qwen2.5 7B, Attacks=Phrase2026.03 | 93.15 | 79.43 | |
| PPLModel=Qwen2.5 7B, Attacks=Word2026.03 | 92.82 | 34.41 | |
| CleanGenModel=Qwen2.5 7B, Attacks=Joint2026.03 | 92.05 | 78.62 | |
| CleanGenModel=Qwen2.5 14B, Attacks=System2026.03 | 87 | 73.55 | |
| PPLModel=Qwen2.5 7B, Attacks=Sentence2026.03 | 80 | 33.6 | |
| ParagraphModel=Qwen2.5 7B, Attacks=Joint2026.03 | 74.05 | 51.2 | |
| ONIONModel=Qwen2.5 7B, Attacks=Word2026.03 | 63.22 | 0.4 | |
| ParagraphModel=Qwen2.5 14B, Attacks=System2026.03 | 59.2 | 28.96 | |
| STRIPModel=Qwen2.5 14B, Attacks=System2026.03 | 56.8 | 62.4 | |
| ONIONModel=Qwen2.5 7B, Attacks=Joint2026.03 | 45.6 | 0.8 | |
| ParagraphModel=Qwen2.5 7B, Attacks=Sentence2026.03 | 22.66 | 48.17 | |
| ParagraphModel=Qwen2.5 7B, Attacks=Phrase2026.03 | 20.09 | 44.35 | |
| ParagraphModel=Qwen2.5 7B, Attacks=Word2026.03 | 15.69 | 39.67 | |
| ONIONModel=Qwen2.5 7B, Attacks=Phrase2026.03 | 3.65 | 0.8 | |
| ONIONModel=Qwen2.5 7B, Attacks=Sentence2026.03 | 1.33 | 0.4 | |
| ONIONModel=Qwen2.5 14B, Attacks=System2026.03 | 0.4 | 1 |