Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Pairwise Preference Ranking on UltraFeedback 5% holdout (test)

87Pairwise Accuracy (RM1-Honest)

BENCHALIGN

70.46474.75779.0583.343Feb 2, 2026
Updated 4d ago

Evaluation Results

MethodLinks
2026.02
8786.70.9050.899
2026.02
71.871.50.6270.611
2026.02
71.570.60.6140.594
2026.02
71.170.90.6080.593