Safety Defense on Beavertails
10ASRSPARD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SPARDBackbone=Qwen-2.5-7B-Instruct2026.05 | 10 | 1.32 | |
| LisaBackbone=Qwen-2.5-7B-Instruct2026.05 | 24.4 | 1.73 | |
| Qwen-2.5-7B-InstructBackbone=Qwen-2.5-7B-Instruct2026.05 | 25.8 | 1.71 | |
| SafeGradBackbone=Qwen-2.5-7B-Instruct2026.05 | 32 | 1.97 | |
| PTSTBackbone=Qwen-2.5-7B-Instruct2026.05 | 62.4 | 3.03 | |
| SafeInstrBackbone=Qwen-2.5-7B-Instruct2026.05 | 65.6 | 3.21 | |
| SFTBackbone=Qwen-2.5-7B-Instruct2026.05 | 83.6 | 3.92 |