LLM Jailbreaking on AdvBench (ASR-M & ASR-J)
88.27ASR-MPEO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PEOModel=Llama-3.2-3B-Instruct2026.04 | 88.27 | 75.58 | |
| SPTModel=Llama-3.2-3B-Instruct2026.04 | 82.69 | 18.08 | |
| nanoGCGModel=Vicuna-7B v1.32026.04 | 81.35 | 29.04 | |
| PEOModel=Vicuna-7B v1.32026.04 | 70.58 | 78.65 | |
| SPTModel=Vicuna-7B v1.32026.04 | 69.81 | 46.15 | |
| PEOModel=Llama-2-7B-Chat2026.04 | 59.62 | 71.73 | |
| BEASTModel=Llama-3.2-3B-Instruct2026.04 | 58.08 | 15.19 | |
| nanoGCGModel=Llama-2-7B-Chat2026.04 | 55.38 | 37.5 | |
| nanoGCGModel=Llama-3.2-3B-Instruct2026.04 | 48.46 | 23.46 | |
| SPTModel=Llama-2-7B-Chat2026.04 | 40.58 | 49.23 | |
| PEOModel=Qwen3-4B-Instruct2026.04 | 38.65 | 72.69 | |
| BEASTModel=Qwen3-4B-Instruct2026.04 | 28.08 | 4.62 | |
| BEASTModel=Vicuna-7B v1.32026.04 | 24.42 | 24.62 | |
| SPTModel=Qwen3-4B-Instruct2026.04 | 18.85 | 18.27 | |
| nanoGCGModel=Qwen3-4B-Instruct2026.04 | 15.38 | 8.46 | |
| BEASTModel=Llama-2-7B-Chat2026.04 | 10.38 | 5.58 |