Jailbreaking on AdvBench 20% evaluation
97.12ASRA-LQR+
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| A-LQR+Model=Qwen2.5-14B Instruct2026.04 | 97.12 | 0 | |
| A-LQR+Model=Qwen2.5-3B Instruct2026.04 | 96.15 | 0 | |
| AASModel=Llama-3.2-3B Instruct2026.04 | 91.35 | 0 | |
| A-LQR+Model=Llama-3.2-3B Instruct2026.04 | 91.35 | 0 | |
| A-LQR+Model=Llama-3.1-8B Instruct2026.04 | 91.35 | 0 | |
| AASModel=Llama-3.1-8B Instruct2026.04 | 90.38 | 0 | |
| S-PIDModel=Llama-3.1-8B Instruct2026.04 | 87.5 | 0.9615 | |
| A-LQRModel=Qwen2.5-3B Instruct2026.04 | 85.58 | 0 | |
| A-LQRModel=Llama-3.1-8B Instruct2026.04 | 85.58 | 0 | |
| S-PIDModel=Qwen2.5-3B Instruct2026.04 | 83.65 | 0 | |
| S-PIDModel=Llama-3.2-3B Instruct2026.04 | 79.81 | 0.9615 | |
| AASModel=Qwen2.5-14B Instruct2026.04 | 77.88 | 0 | |
| A-LQR+Model=Gemma-2-9B Instruct2026.04 | 76.92 | 0 | |
| AASModel=Qwen2.5-3B Instruct2026.04 | 75.96 | 0 | |
| A-LQRModel=Llama-3.2-3B Instruct2026.04 | 71.15 | 0 | |
| A-LQRModel=Qwen2.5-14B Instruct2026.04 | 69.23 | 0 | |
| S-PIDModel=Qwen2.5-14B Instruct2026.04 | 65.38 | 0 | |
| S-PIDModel=Gemma-2-9B Instruct2026.04 | 16.35 | 0 | |
| A-LQRModel=Gemma-2-9B Instruct2026.04 | 14.42 | 0 | |
| AASModel=Gemma-2-9B Instruct2026.04 | 12.5 | 68.27 | |
| OriginalModel=Llama-3.2-3B Instruct2026.04 | 4.808 | 81.73 | |
| OriginalModel=Llama-3.1-8B Instruct2026.04 | 3.846 | 96.15 | |
| OriginalModel=Qwen2.5-14B Instruct2026.04 | 1.923 | 97.12 | |
| OriginalModel=Qwen2.5-3B Instruct2026.04 | 0.9615 | 98.08 | |
| OriginalModel=Gemma-2-9B Instruct2026.04 | 0 | 100 |