Pairwise Preference Ranking on Helpsteer 5% holdout (test)
84.9Pairwise Accuracy (RM1-Helpful)
BENCHALIGN
Updated 4d ago
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
2026.02 | 84.9 | 79.2 | 0.866 | 0.762 | |
2026.02 | 70.8 | 60.4 | 0.581 | 0.312 | |
2026.02 | 70.4 | 60.1 | 0.569 | 0.307 | |
2026.02 | 67 | 58.9 | 0.384 | 0.257 |