Response Evaluation on UltraFeedback (test)
18.8Win RateMGDA-DECOUPLED
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MGDA-DECOUPLEDAlgorithm=MGDA-DECOUPLED2026.04 | 18.8 | 3 | 78.2 | |
| MGDA-NORMALISEDAlgorithm=MGDA-NORMALISED2026.04 | 18.5 | 2.9 | 78.6 | |
| UNIFORMAlgorithm=UNIFORM2026.04 | 18.5 | 3 | 78.6 | |
| GROUPDROAlgorithm=GROUPDRO2026.04 | 18.2 | 2.7 | 79.1 | |
| CDPOAlgorithm=CDPO2026.04 | 17.5 | 2.7 | 79.8 | |
| REFERENCEAlgorithm=REFERENCE2026.04 | 16.6 | 2.7 | 80.7 |