Detection Efficiency on Alpaca OnlyTarget Long (malicious)
8.378ATGRParagraph
Evaluation Results
| Method | Links | |
|---|---|---|
| ParagraphModel=Qwen2.5 14B, Attack=Ignore2026.03 | 8.378 | |
| ParagraphModel=Qwen2.5 14B, Attack=System2026.03 | 8.016 | |
| ParagraphModel=Qwen2.5 14B, Attack=Tree2026.03 | 7.568 | |
| ParagraphModel=Qwen2.5 14B, Attack=Complete2026.03 | 6.998 | |
| ParagraphModel=Qwen2.5 7B, Attack=Word2026.03 | 6.743 | |
| ParagraphModel=Qwen2.5 7B, Attack=Phrase2026.03 | 6 | |
| ParagraphModel=Qwen2.5 7B, Attack=Joint2026.03 | 5.526 | |
| STRIPModel=Qwen2.5 14B, Attack=Tree2026.03 | 5.353 | |
| ParagraphModel=Qwen2.5 7B, Attack=Sentence2026.03 | 5.181 | |
| STRIPModel=Qwen2.5 14B, Attack=System2026.03 | 4.892 | |
| STRIPModel=Qwen2.5 14B, Attack=Ignore2026.03 | 3.836 | |
| STRIPModel=Qwen2.5 14B, Attack=Complete2026.03 | 3.724 | |
| ONIONModel=Qwen2.5 14B, Attack=Tree2026.03 | 3.555 | |
| ONIONModel=Qwen2.5 14B, Attack=Complete2026.03 | 3.466 | |
| STRIPModel=Qwen2.5 7B, Attack=Word2026.03 | 3.194 | |
| STRIPModel=Qwen2.5 7B, Attack=Joint2026.03 | 2.912 | |
| STRIPModel=Qwen2.5 7B, Attack=Phrase2026.03 | 2.895 | |
| STRIPModel=Qwen2.5 7B, Attack=Sentence2026.03 | 2.864 | |
| ONIONModel=Qwen2.5 7B, Attack=Phrase2026.03 | 2.386 | |
| ONIONModel=Qwen2.5 14B, Attack=Ignore2026.03 | 2.303 | |
| ONIONModel=Qwen2.5 7B, Attack=Word2026.03 | 2.108 | |
| ONIONModel=Qwen2.5 7B, Attack=Joint2026.03 | 2.09 | |
| ONIONModel=Qwen2.5 14B, Attack=System2026.03 | 1.649 | |
| ONIONModel=Qwen2.5 7B, Attack=Sentence2026.03 | 1.642 | |
| CleanGenModel=Qwen2.5 14B, Attack=System2026.03 | 1.082 | |
| CleanGenModel=Qwen2.5 14B, Attack=Complete2026.03 | 1.079 | |
| CleanGenModel=Qwen2.5 14B, Attack=Tree2026.03 | 1.074 | |
| PPLModel=Qwen2.5 7B, Attack=Sentence2026.03 | 1.071 | |
| PPLModel=Qwen2.5 14B, Attack=System2026.03 | 1.069 | |
| PPLModel=Qwen2.5 14B, Attack=Ignore2026.03 | 1.066 | |
| CleanGenModel=Qwen2.5 14B, Attack=Ignore2026.03 | 1.061 | |
| CleanGenModel=Qwen2.5 7B, Attack=Word2026.03 | 1.054 | |
| DualSentinelModel=Qwen2.5 14B, Attack=System2026.03 | 1.052 | |
| CleanGenModel=Qwen2.5 7B, Attack=Sentence2026.03 | 1.045 | |
| ConfGuardModel=Qwen2.5 7B, Attack=Joint2026.03 | 1.041 | |
| PPLModel=Qwen2.5 7B, Attack=Word2026.03 | 1.038 | |
| PPLModel=Qwen2.5 14B, Attack=Tree2026.03 | 1.029 | |
| DualSentinelModel=Qwen2.5 14B, Attack=Tree2026.03 | 1.029 | |
| PPLModel=Qwen2.5 14B, Attack=Complete2026.03 | 1.028 | |
| CleanGenModel=Qwen2.5 7B, Attack=Joint2026.03 | 1.021 | |
| ConfGuardModel=Qwen2.5 7B, Attack=Word2026.03 | 1.019 | |
| ConfGuardModel=Qwen2.5 7B, Attack=Sentence2026.03 | 1.019 | |
| ConfGuardModel=Qwen2.5 14B, Attack=System2026.03 | 1.01 | |
| DualSentinelModel=Qwen2.5 7B, Attack=Sentence2026.03 | 1.009 | |
| ConfGuardModel=Qwen2.5 14B, Attack=Ignore2026.03 | 1.007 | |
| ConfGuardModel=Qwen2.5 14B, Attack=Complete2026.03 | 1.005 | |
| DualSentinelModel=Qwen2.5 14B, Attack=Complete2026.03 | 1.003 | |
| CleanGenModel=Qwen2.5 7B, Attack=Phrase2026.03 | 1.003 | |
| PPLModel=Qwen2.5 7B, Attack=Joint2026.03 | 1.001 | |
| ConfGuardModel=Qwen2.5 7B, Attack=Phrase2026.03 | 0.982 | |
| DualSentinelModel=Qwen2.5 7B, Attack=Joint2026.03 | 0.972 | |
| DualSentinelModel=Qwen2.5 7B, Attack=Word2026.03 | 0.956 | |
| PPLModel=Qwen2.5 7B, Attack=Phrase2026.03 | 0.951 | |
| DualSentinelModel=Qwen2.5 7B, Attack=Phrase2026.03 | 0.944 | |
| ConfGuardModel=Qwen2.5 14B, Attack=Tree2026.03 | 0.902 | |
| DualSentinelModel=Qwen2.5 14B, Attack=Ignore2026.03 | 0.812 |