List your company

LLM Alignment on UltraFeedback (in-domain)

80.6Win Rate (KL, alpha=1)GEB-π
72.581674.663376.74578.8267Sep 27, 2025

Evaluation results

MethodLinks
GEB-πBase Model=Mistral-Instruct-v0.32025.0980.60.753174.410.672351.750.5986
GEB-1/πBase Model=Mistral-Instruct-v0.32025.0979.690.761473.870.634251.860.6088
GEB-arctanh(π − 1)Base Model=Mistral-Instruct-v0.32025.0979.60.754474.570.664251.690.5962
f-DPOBase Model=Mistral-Instruct-v0.32025.0977.490.742672.70.648850.830.5902
GEB-arctanh(π − 1)Base Model=LLaMA-3-8B-SFT2025.0974.550.803272.570.790268.340.7612
GEB-1/πBase Model=LLaMA-3-8B-SFT2025.0974.360.808472.280.773367.990.7618
GEB-πBase Model=LLaMA-3-8B-SFT2025.0973.90.804872.080.792167.790.7698
f-DPOBase Model=LLaMA-3-8B-SFT2025.0972.890.792671.190.791566.620.7566