Jailbreak Attack on AdvBench (Loss, ASR, Avg ASR)
0.16LossFaster-GCG++
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Faster-GCG++Target LLM=Qwen3.5-4B2024.10 | 0.16 | 86.2 | — | |
| Faster-GCG++Target LLM=Llama2-7B2024.10 | 0.177 | 60.4 | — | |
| Faster-GCGTarget LLM=Qwen3.5-4B2024.10 | 0.259 | 79.8 | — | |
| TAOTarget LLM=Qwen3.5-4B2024.10 | 0.312 | 77.9 | — | |
| Faster-GCGTarget LLM=Llama2-7B2024.10 | 0.315 | 55.2 | — | |
| TAOTarget LLM=Llama2-7B2024.10 | 0.351 | 56.7 | — | |
| GCGTarget LLM=Llama2-7B2024.10 | 0.36 | 49.7 | — | |
| GCGTarget LLM=Qwen3.5-4B2024.10 | 0.361 | 68.9 | — | |
| Faster-GCG++Target LLM=Llama3.1-8B2024.10 | 0.489 | 63.1 | — | |
| Faster-GCG++Target LLM=Qwen3-4B2024.10 | 0.592 | 83.9 | — | |
| Faster-GCGTarget LLM=Llama3.1-8B2024.10 | 0.723 | 49.6 | — | |
| TAOTarget LLM=Llama3.1-8B2024.10 | 0.751 | 51.5 | — | |
| GCGTarget LLM=Llama3.1-8B2024.10 | 0.796 | 45.9 | — | |
| Faster-GCGTarget LLM=Qwen3-4B2024.10 | 0.826 | 51.5 | — | |
| TAOTarget LLM=Qwen3-4B2024.10 | 0.901 | 47.8 | — | |
| GCGTarget LLM=Qwen3-4B2024.10 | 0.906 | 49.1 | — | |
| Faster-GCGTarget LLM=Overall2024.10 | — | — | 59 | |
| Faster-GCG++Target LLM=Overall2024.10 | — | — | 73.4 | |
| GCGTarget LLM=Overall2024.10 | — | — | 53.4 | |
| TAOTarget LLM=Overall2024.10 | — | — | 58.5 |