Defense Robustness on AutoDAN Attack
100Keyword Success RateES2
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ES2LLM=*Llama-2 (7B-Chat)2026.03 | 100 | 100 | 94 | |
| ES2LLM=Llama-3 (8B-Instruct)2026.03 | 100 | 100 | 100 | |
| ES2LLM=Qwen-2.5 (7B-Instruct)2026.03 | 99 | 94 | 94 | |
| STLLLM=Llama-3 (8B-Instruct)2026.03 | 98 | 98 | 97 | |
| DPLLLM=Llama-3 (8B-Instruct)2026.03 | 97 | 97 | 96 | |
| Base ModelLLM=Llama-3 (8B-Instruct)2026.03 | 95 | 95 | 86 | |
| DPLLLM=Qwen-2.5 (7B-Instruct)2026.03 | 92 | 86 | 86 | |
| STLLLM=*Llama-2 (7B-Chat)2026.03 | 91 | 87 | 78 | |
| DPLLLM=*Llama-2 (7B-Chat)2026.03 | 90 | 87 | 76 | |
| STLLLM=Qwen-2.5 (7B-Instruct)2026.03 | 89 | 84 | 77 | |
| Base ModelLLM=Qwen-2.5 (7B-Instruct)2026.03 | 81 | 72 | 72 | |
| Base ModelLLM=*Llama-2 (7B-Chat)2026.03 | 80 | 77 | 72 | |
| ES2LLM=Mistral (7B-Instruct)2026.03 | 60 | 54 | 47 | |
| DPLLLM=Mistral (7B-Instruct)2026.03 | 20 | 17 | 16 | |
| STLLLM=Mistral (7B-Instruct)2026.03 | 18 | 13 | 10 | |
| Base ModelLLM=Mistral (7B-Instruct)2026.03 | 10 | 4 | 4 |