List your company

Generative Performance on Ultrafeedback 61.1k (test)

69.8Win RateFA-DPO
50.45655.47860.565.522Nov 30, 2025

Evaluation results

MethodLinks
FA-DPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.1169.8
FA-DPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.1169.8
FA-DPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.1168.45
cDPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.1167.8
ROPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.1167.25
cDPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.1167.05
DPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.1167
FA-DPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.1166.9
FA-DPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.1166.85
ROPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.1166.85
ROPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.1166.8
cDPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.1166.65
cDPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.1166.55
cDPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.1166.05
rDPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.1165.6
ROPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.1165.55
DPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.1164.7
ROPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.1164.4
DPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.1164.35
SIMPOFlipping ratio (η)=0%, Model Backbone=Pythia-1B2025.1162.6
DPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.1160.1
rDPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.1157.85
rDPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.1157.8
rDPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.1156.65
SIMPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.1156.5
DPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.1156.35
SIMPOFlipping ratio (η)=40%, Model Backbone=Pythia-1B2025.1155.8
SIMPOFlipping ratio (η)=30%, Model Backbone=Pythia-1B2025.1154.9
rDPOFlipping ratio (η)=20%, Model Backbone=Pythia-1B2025.1154.85
SIMPOFlipping ratio (η)=10%, Model Backbone=Pythia-1B2025.1151.2