List your company

Harmlessness Evaluation on PKU-SafeRLHF-30K

87.25Win RateSafeDPO
56.705264.635172.56580.4949May 26, 2025

Evaluation results

MethodLinks
SafeDPOBaseline=DPO-HELPFUL2025.0587.253.759
SafeDPOBaseline=DPO-SAFEBETTER2025.0585.983.7910.23
SafeDPOBaseline=DPO-HARMLESS2025.0579.754.515.75
SafeDPOBaseline=SACPO2025.0574.755.7519.5
SafeDPOBaseline=P-SACPO2025.05726.521.5
SafeDPOBaseline=SafeRLHF2025.0557.885.7536.38