List your company

LLM Alignment on Ultrafeedback

86Win RateGreedy SFT
21.915238.552655.1971.8274Sep 30, 2025Nov 11, 2025Dec 23, 2025Feb 3, 2026Mar 17, 2026Apr 28, 2026Jun 10, 2026

Evaluation results

MethodLinks
Greedy SFTBase Model=Llama 3B2025.09860.58
Greedy DPOBase Model=Llama 3B2025.09860.68
Bo2Base Model=Llama 3B2025.09850.85
Greedy SFTBase Model=Llama 8B2025.09850.87
Bo2Base Model=Llama 8B2025.09851.06
Greedy DPOBase Model=Llama 8B2025.09840.98
EFTBase Model=Llama 3B2025.09831.04
EFTBase Model=Llama 8B2025.09811.27
Greedy SFTBase Model=Qwen 0.6B2025.0980-0.88
Greedy SFTBase Model=Qwen 4B2025.09800.22
Greedy DPOBase Model=Qwen 4B2025.09790.29
Greedy DPOBase Model=Qwen 0.6B2025.0978-0.69
Bo2Base Model=Qwen 0.6B2025.0978-0.62
Bo2Base Model=Qwen 4B2025.09770.47
EFTBase Model=Qwen 4B2025.09730.58
EFTBase Model=Qwen 0.6B2025.0967-0.19
Clean Data2026.0627.15—
DeMixrepair=true2026.0626.04—
DeMixrepair=false2026.0626.02—
DeMix-unifrepair=false2026.0625.97—
DeMix-unifrepair=true2026.0625.84—
DDA-repair2026.0625.82—
DDA-select2026.0625.37—
Error. Data2026.0624.38—
AADBase Model=Llama 3B2025.09—2.21
AADBase Model=Llama 8B2025.09—2.22
AADBase Model=Qwen 0.6B2025.09—0.34
AADBase Model=Qwen 4B2025.09—1.19