Adversarial Robustness on PAIR
26ASRNo Defense
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| No DefenseModel=Llama2-7b2026.02 | 26 | 2.36 | |
| SD_TinyExpertModel=Llama2-7b2026.02 | 20 | 2.12 | |
| No DefenseModel=Llama2-13b2026.02 | 18 | 2.08 | |
| ParaphraseModel=Llama2-7b2026.02 | 16 | 2.32 | |
| SD_DaExpertModel=Llama2-13b2026.02 | 16 | 1.94 | |
| SSDModel=Llama2-7b2026.02 | 14 | 2.06 | |
| SSDModel=Llama2-13b2026.02 | 14 | 1.86 | |
| Self-ExamModel=Llama2-13b2026.02 | 12 | 1.78 | |
| ParaphraseModel=Llama2-13b2026.02 | 10 | 2.4 | |
| SD_TinyExpertModel=Llama2-13b2026.02 | 10 | 1.82 | |
| SafeDecodingModel=Llama2-7b2026.02 | 6 | 1.6 | |
| SafeDecodingModel=Llama2-13b2026.02 | 6 | 1.66 | |
| SD_DaExpertModel=Llama2-7b2026.02 | 4 | 1.36 | |
| Self-ReminderModel=Llama2-7b2026.02 | 2 | 1.18 | |
| NGSDModel=Llama2-7b2026.02 | 2 | 1.6 | |
| Self-ReminderModel=Llama2-13b2026.02 | 2 | 1.2 | |
| NGSDModel=Llama2-13b2026.02 | 2 | 1.56 | |
| Self-ExamModel=Llama2-7b2026.02 | 0 | 1 |