Preference Optimization on HelpSteer2 (test)
0.8Avg Pref Score vs QWEN2.5-0.5BSTACKELBERGGDA
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| STACKELBERGGDARole=Follower2025.12 | 0.8 | 0.656 | 0.594 | 0.605 | 0 | |
| STACKELBERGGDARole=Leader2025.12 | 0.734 | 0.613 | 0.503 | 0 | 0.395 | |
| NASH-MD-PG2025.12 | 0.721 | 0.607 | 0 | 0.497 | 0.406 | |
| RLOO2025.12 | 0.593 | 0 | 0.393 | 0.387 | 0.344 | |
| QWEN2.5-0.5B2025.12 | 0 | 0.407 | 0.279 | 0.266 | 0.2 |