Adversarial Attack on 1,000-pair benchmark Gemma-3-27B-it
24.6ASRM-Attack-V2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| M-Attack-V2Similarity Threshold=0.72026.05 | 24.6 | 39.6 | |
| FRA-AttackSimilarity Threshold=0.72026.05 | 20.8 | 36.9 | |
| FOA-AttackSimilarity Threshold=0.72026.05 | 16.6 | 33.9 | |
| M-AttackSimilarity Threshold=0.72026.05 | 14.6 | 31.8 | |
| AttackVLMSimilarity Threshold=0.72026.05 | 14 | 25.8 | |
| SSA-CWASimilarity Threshold=0.72026.05 | 12 | 27.2 | |
| AdvDiffVLMSimilarity Threshold=0.72026.05 | 10 | 26.6 | |
| AnyAttackSimilarity Threshold=0.72026.05 | 7 | 21.6 |