Safety Evaluation on Prefilling Attacks Qwen2.5-14B-Instruct
11.5ASRSafeThinker
Evaluation Results
| Method | Links | |
|---|---|---|
| SafeThinkerPrefix length=102026.01 | 11.5 | |
| SafeThinkerPrefix length=402026.01 | 12.4 | |
| SafeThinkerPrefix length=202026.01 | 13.6 | |
| No DefensePrefix length=102026.01 | 32.1 | |
| No DefensePrefix length=202026.01 | 32.1 | |
| SafeDecodingPrefix length=202026.01 | 32.1 | |
| SafeDecodingPrefix length=402026.01 | 32.7 | |
| No DefensePrefix length=402026.01 | 33 | |
| SafeDecodingPrefix length=102026.01 | 33 |