Reward Modeling Evaluation on HelpSteer3 (test)
-5.89ScoreDPO+Filter
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DPO+FilterPη Type=12025.10 | -5.89 | 73 | |
| DPOPη Type=12025.10 | -6.56 | 67 | |
| DPO+FilterPη Type=22025.10 | -6.83 | 68 | |
| DPOPη Type=22025.10 | -6.91 | 67 | |
| BasePη Type=-2025.10 | -8.39 | 50 |