Reward Modeling on RewardBench Filtered (test)
98.3ChatBradley-Terry (BT)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Bradley-Terry (BT)training=standard2025.12 | 98.3 | 71.4 | 83.1 | 85.1 | |
| Format Decoupling (FD)2025.12 | 97.2 | 72.8 | 82.9 | 89.7 | |
| DIR2025.12 | 93 | 80.1 | 89.6 | 92.2 | |
| Bradley-Terry (BT†)training=filtered (removing format-biased examples)2025.12 | 92.2 | 64.4 | 75.5 | 81.4 |