Bias Scoring Correlation on RealToxicityPrompts sampled completions
0.4748Spearman Correlation CoefficientCustom
Evaluation Results
| Method | Links | |
|---|---|---|
| CustomReward Model=Custom2026.06 | 0.4748 | |
| Stereotype RoBERTaReward Model=Stereotype RoBERTa2026.06 | 0.3769 | |
| DetoxifyReward Model=Detoxify2026.06 | 0.1812 | |
| RoBERTa Hate SpeechReward Model=RoBERTa Hate Speech2026.06 | 0.1701 | |
| DistilRoBERTa-Bias MBICReward Model=DistilRoBERTa-Bias MBIC2026.06 | 0.1427 | |
| DistilRoBERTa-BiasReward Model=DistilRoBERTa-Bias2026.06 | 0.0911 |