Jailbreak Attack on 50 prompts Vicuna victim
48Positive Exits CountLLaMA-Guard-3-8B + CMPE
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LLaMA-Guard-3-8B + CMPEVictim=Vicuna, Attacker=GPTFuzz, Total prompts=502026.06 | 48 | 2 | 44 | 4 | 0 | 0 | 6.8 | |
| PAIRDefense=LLaMA-Guard-3-8B + CMPE, Total=502026.06 | 43 | 7 | 43 | 0 | 0 | 0 | 20.4 | |
| LLaMA-Guard-3-8BVictim=Vicuna, Attacker=GPTFuzz, Total prompts=502026.06 | 23 | 27 | 0 | 13 | 10 | 20 | 37.5 | |
| PAIRDefense=LLaMA-Guard-3-8B, Total=502026.06 | 8 | 42 | 0 | 3 | 5 | 10 | 45.4 |