List your company
ResearchFormerly WizWand

Preference Alignment Evaluation on Nectar

3.7Reward (R)AAD
-0.94880.25811.4652.6719Sep 30, 2025
Updated 4mo ago

Evaluation results

MethodLinks
AADBase Model=Llama 8B2025.093.7———
AADBase Model=Llama 3B2025.093.63———
AADModel=Llama 3B, Training Protocol=full finetuning2025.093.45———
EFTModel=Llama 3B, Training Protocol=full finetuning2025.093.35——58
EFTBase Model=Llama 8B2025.093.3——75
AADModel=Llama 1B, Training Protocol=full finetuning2025.093.05———
Bo2Model=Llama 3B, Training Protocol=full finetuning2025.092.9——79
AADBase Model=Qwen 4B2025.092.71———
EFTModel=Llama 1B, Training Protocol=full finetuning2025.092.68——65
Bo2Base Model=Llama 8B2025.092.64——93
Greedy DPOModel=Llama 3B, Training Protocol=full finetuning2025.092.46——89
EFTBase Model=Qwen 4B2025.092.35——65
Bo2Model=Llama 1B, Training Protocol=full finetuning2025.092.28——77
EFTBase Model=Llama 3B2025.092.28——89
Bo2Base Model=Llama 3B2025.092.15——95
Greedy DPOBase Model=Llama 8B2025.092.12——99
Bo2Base Model=Qwen 4B2025.091.99——74
AADBase Model=Qwen 0.6B2025.091.68———
Greedy DPOBase Model=Llama 3B2025.091.45——98
Greedy DPOBase Model=Qwen 4B2025.091.45——85
Greedy DPOModel=Llama 1B, Training Protocol=full finetuning2025.091.32——91
EFTBase Model=Qwen 0.6B2025.091.23——58
Greedy SFTBase Model=Llama 8B2025.091.17——99
Bo2Base Model=Qwen 0.6B2025.091.07——70
Greedy SFTBase Model=Qwen 4B2025.090.77——94
Greedy SFTModel=Llama 3B, Training Protocol=full finetuning2025.090.72——98
Greedy SFTBase Model=Llama 3B2025.090.72——99
Greedy DPOBase Model=Qwen 0.6B2025.090.09——84
Greedy SFTModel=Llama 1B, Training Protocol=full finetuning2025.09-0.26——98
Greedy SFTBase Model=Qwen 0.6B2025.09-0.77——93
Pref-CTRLTrained Dataset=HH-RLHF, Zero-shot transfer setting=true2026.04—32.3335.67—
RE-ControlTrained Dataset=HH-RLHF, Zero-shot transfer setting=true2026.04—30.6733.67—