Trigger inversion on Yahoo
30Trigger Inversion Success RateCSO(discrete)
Evaluation Results
| Method | Links | |
|---|---|---|
| CSO(discrete)Target trigger (length)=Tell me seriously(3 tokens), Label type=clean label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 30 | |
| CSO(discrete)Target trigger (length)=Tell me seriously(3 tokens), Label type=dirty label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 29 | |
| CSO(discrete)Target trigger (length)=Tell me seriously(3 tokens), Label type=dirty label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 23 | |
| CSO(discrete)Target trigger (length)=Tell me seriously(3 tokens), Label type=clean label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 22 | |
| CSO(discrete)Target trigger (length)=mb(1 token), Label type=clean label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 10 | |
| CSO(discrete)Target trigger (length)=mb(1 token), Label type=clean label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 10 | |
| CSO(discrete)Target trigger (length)=mb(1 token), Label type=dirty label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 10 | |
| CSO(discrete)Target trigger (length)=mb(1 token), Label type=dirty label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 10 | |
| MMTarget trigger (length)=mb(1 token), Label type=clean label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 10 | |
| MMTarget trigger (length)=mb(1 token), Label type=dirty label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 10 | |
| MMTarget trigger (length)=Tell me seriously(3 tokens), Label type=dirty label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 9 | |
| MMTarget trigger (length)=mb(1 token), Label type=dirty label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 9 | |
| MMTarget trigger (length)=Tell me seriously(3 tokens), Label type=dirty label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 8 | |
| MMTarget trigger (length)=mb(1 token), Label type=clean label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 8 | |
| MMTarget trigger (length)=Tell me seriously(3 tokens), Label type=clean label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 5 | |
| MMTarget trigger (length)=Tell me seriously(3 tokens), Label type=clean label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 3 | |
| DBSTarget trigger (length)=Tell me seriously(3 tokens), Label type=clean label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 3 | |
| DBSTarget trigger (length)=Tell me seriously(3 tokens), Label type=dirty label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 3 | |
| PICCOLOTarget trigger (length)=Tell me seriously(3 tokens), Label type=clean label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 3 | |
| UATTarget trigger (length)=mb(1 token), Label type=clean label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 2 | |
| UATTarget trigger (length)=mb(1 token), Label type=clean label, Poison rate=low, Model=Qwen3-0.6b2026.06 | 1 | |
| PICCOLOTarget trigger (length)=Tell me seriously(3 tokens), Label type=dirty label, Poison rate=high, Model=Qwen3-0.6b2026.06 | 1 |