Safety Evaluation on SorryBench
8.22ASRDb as Alpaca
Evaluation results
| Method | Links | |
|---|---|---|
| Db as AlpacaSetting / Model=RLVR, Llama3-U2026.03 | 8.22 | |
| Db as AlpacaSetting / Model=P-SFT, Llama2, ASR evaluation protocol=prefill attacks2026.03 | 18.22 | |
| Db as Our DataSetting / Model=P-SFT, Llama2, ASR evaluation protocol=prefill attacks2026.03 | 25.11 | |
| Db as Our DataSetting / Model=RLVR, Llama3-U2026.03 | 25.33 | |
| BaselineSetting / Model=P-SFT, Llama2, ASR evaluation protocol=prefill attacks2026.03 | 47.56 | |
| BaselineSetting / Model=RLVR, Llama3-U2026.03 | 85.71 |