Adversarial Attack Success on AdvBench (sample of 500)
16.5ASRSR-Scout-30B
Evaluation Results
| Method | Links | |
|---|---|---|
| SR-Scout-30BAttacker=HB-FT-LLaMA2-13B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 16.5 | |
| LLaMA-Guard-3-8BAttacker=HB-FT-LLaMA2-13B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 12.6 | |
| SR-Scout-30BAttacker=SR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 12.4 | |
| LLaMA-Guard-3-8BAttacker=SR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 9.8 | |
| SR-Scout-30BAttacker=PAIR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 9.4 | |
| SR-Scout-30BAttacker=GPTFuzz-RoBERTa, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 8.4 | |
| HB-FT-LLaMA2-13BAttacker=SR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 7.7 | |
| LLaMA-Guard-3-8BAttacker=SR-Scout-30B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 7.5 | |
| SR-Scout-30BAttacker=LLaMA-Guard-3-8B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 7.4 | |
| HB-FT-LLaMA2-13BAttacker=SR-Scout-30B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 6.4 | |
| LLaMA-Guard-3-8BAttacker=PAIR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 6.3 | |
| HB-FT-LLaMA2-13BAttacker=PAIR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 5.2 | |
| SR-Scout-30BAttacker=SR-Scout-30B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 4.9 | |
| HB-FT-LLaMA2-13BAttacker=LLaMA-Guard-3-8B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 4.6 | |
| HB-FT-LLaMA2-13BAttacker=GPTFuzz-RoBERTa, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 3.9 | |
| LLaMA-Guard-3-8BAttacker=GPTFuzz-RoBERTa, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 3.9 | |
| PAIR-OSS-120BAttacker=SR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 2.8 | |
| PAIR-OSS-120BAttacker=HB-FT-LLaMA2-13B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 2.5 | |
| PAIR-OSS-120BAttacker=SR-Scout-30B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 2.3 | |
| PAIR-OSS-120BAttacker=GPTFuzz-RoBERTa, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 1.7 | |
| SR-OSS-120BAttacker=HB-FT-LLaMA2-13B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.7 | |
| PAIR-OSS-120BAttacker=PAIR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.7 | |
| SR-OSS-120BAttacker=SR-Scout-30B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.5 | |
| SR-OSS-120BAttacker=PAIR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.5 | |
| SR-OSS-120BAttacker=GPTFuzz-RoBERTa, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.5 | |
| PAIR-OSS-120BAttacker=LLaMA-Guard-3-8B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.5 | |
| SR-OSS-120BAttacker=SR-OSS-120B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.4 | |
| SR-OSS-120BAttacker=LLaMA-Guard-3-8B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.3 | |
| HB-FT-LLaMA2-13BAttacker=HB-FT-LLaMA2-13B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0.1 | |
| LLaMA-Guard-3-8BAttacker=LLaMA-Guard-3-8B, Strategy=CMPE-based detect-and-misdirect, N=100, K=12026.06 | 0 |