Safety Evaluation on ALERT (test)
0.2ASRSafeDecoding
Evaluation Results
| Method | Links | |
|---|---|---|
| SafeDecodingBackbone=Qwen2.5-14B-Instruct2026.01 | 0.2 | |
| SafeThinkerBackbone=Qwen2.5-14B-Instruct2026.01 | 0.2 | |
| Self-ReminderBackbone=Qwen2.5-14B-Instruct2026.01 | 0.6 | |
| PPLBackbone=Qwen2.5-14B-Instruct2026.01 | 1.8 | |
| Self-ExaminationBackbone=Qwen2.5-14B-Instruct2026.01 | 2 | |
| No DefenseBackbone=Qwen2.5-14B-Instruct2026.01 | 2.4 | |
| ICDBackbone=Qwen2.5-14B-Instruct2026.01 | 8 |