Backdoor Attack Generalization on OOD Triggers (test)
84.8ASR (OOD)GREAT
Evaluation Results
| Method | Links | |
|---|---|---|
| GREATalpha=10%, K=2000, Victim Model=Llama-2-7B2025.10 | 84.8 | |
| GREATalpha=10%, K=2000, Victim Model=Llama-3.2-3B2025.10 | 80.5 | |
| GREATalpha=10%, K=2000, Victim Model=Llama-3.2-1B2025.10 | 79.3 | |
| GREATalpha=1%, K=100, Victim Model=Llama-3.2-3B2025.10 | 70 | |
| GREATalpha=1%, K=100, Victim Model=Llama-2-7B2025.10 | 65.2 | |
| Randomalpha=10%, K=2000, Victim Model=Llama-2-7B2025.10 | 65.2 | |
| Randomalpha=1%, K=100, Victim Model=Llama-2-7B2025.10 | 61.5 | |
| Randomalpha=10%, K=2000, Victim Model=Llama-3.2-3B2025.10 | 57 | |
| Randomalpha=1%, K=100, Victim Model=Llama-3.2-3B2025.10 | 56.5 | |
| GREATalpha=1%, K=100, Victim Model=Llama-3.2-1B2025.10 | 54.3 | |
| SUDOalpha=10%, Victim Model=Llama-2-7B2025.10 | 49.1 | |
| Randomalpha=10%, K=2000, Victim Model=Llama-3.2-1B2025.10 | 46 | |
| SUDOalpha=10%, Victim Model=Llama-3.2-1B2025.10 | 45.8 | |
| SUDOalpha=1%, Victim Model=Llama-2-7B2025.10 | 43.4 | |
| SUDOalpha=10%, Victim Model=Llama-3.2-3B2025.10 | 43.2 | |
| SUDOalpha=1%, Victim Model=Llama-3.2-3B2025.10 | 38.8 | |
| Randomalpha=1%, K=100, Victim Model=Llama-3.2-1B2025.10 | 29.5 | |
| SUDOalpha=1%, Victim Model=Llama-3.2-1B2025.10 | 29.3 |