Overrefusal Evaluation on Koala
4.44Refusal RateDb as Our Data
Evaluation Results
| Method | Links | |
|---|---|---|
| Db as Our DataSetting / Model=RLVR, Llama3-U2026.03 | 4.44 | |
| BaselineSetting / Model=RLVR, Llama3-U2026.03 | 5 | |
| Db as AlpacaSetting / Model=RLVR, Llama3-U2026.03 | 6.67 | |
| BaselineSetting / Model=P-SFT, Llama22026.03 | 7.77 | |
| Db as Our DataSetting / Model=P-SFT, Llama22026.03 | 12.22 | |
| Db as AlpacaSetting / Model=P-SFT, Llama22026.03 | 33.33 |