List your company

Human Preference Alignment on PickScore (Comprehensive Metrics)

23.55PickScore (Task)VGPO
21.646822.140922.63523.1291Dec 13, 2025

Evaluation results

MethodLinks
VGPOKL Regularization=w/o KL2025.1223.555.974.1812823.73.34
Flow-GRPOKL Regularization=w/o KL2025.1223.416.154.1612423.563.33
VGPOKL Regularization=w/ KL2025.1223.415.94.2313223.613.39
Flow-GRPOKL Regularization=w/ KL2025.1223.315.924.2212823.533.38
SD3.5-M2025.1221.725.394.078722.343.07