List your company

Direct Preference Optimization on Skywork AlpacaEval 2.0

20.56LCWRDifficulty-Based Preference Data Selection
1.85046.707711.56516.4223Aug 6, 2025

Evaluation results

MethodLinks
Difficulty-Based Preference Data SelectionSelection Method=Ours2025.0820.5619.38
RandomSelection Method=Random2025.0819.618.57
ZIP†Selection Method=ZIP2025.0818.7418.96
Full SetSelection Method=Full Set2025.0818.1317.54
DiverseEvol†Selection Method=DiverseEvol2025.0817.7518.33
SDPOSelection Method=SDPO2025.0817.4618.56
Tulu3-SFTSelection Method=SFT Baseline2025.082.572.16