List your company
ResearchFormerly WizWand

Preference Alignment on OpenRLHF Mixture

7.6RewardAAD
0.55922.38714.2156.0429Sep 30, 2025
Updated 4mo ago

Evaluation results

MethodLinks
AADBase Model=Llama 8B2025.097.6—
AADBase Model=Llama 3B2025.097.28—
EFTBase Model=Llama 8B2025.096.840.67
AADModel=Llama 3B, Training Protocol=full finetuning2025.096.26—
EFTBase Model=Llama 3B2025.096.180.72
Bo2Model=Llama 3B, Training Protocol=full finetuning2025.095.8857
Bo2Base Model=Llama 8B2025.095.60.85
AADBase Model=Qwen 4B2025.095.45—
Bo2Base Model=Llama 3B2025.095.340.83
EFTBase Model=Qwen 4B2025.095.290.54
EFTModel=Llama 3B, Training Protocol=full finetuning2025.095.2470
Greedy DPOBase Model=Llama 8B2025.094.930.89
Greedy DPOModel=Llama 3B, Training Protocol=full finetuning2025.094.9173
Greedy DPOBase Model=Llama 3B2025.094.540.88
Bo2Base Model=Qwen 4B2025.094.480.69
Bo2Model=Llama 1B, Training Protocol=full finetuning2025.094.0751
AADModel=Llama 1B, Training Protocol=full finetuning2025.094.04—
Greedy SFTBase Model=Llama 8B2025.093.890.93
EFTModel=Llama 1B, Training Protocol=full finetuning2025.093.6457
Greedy SFTModel=Llama 3B, Training Protocol=full finetuning2025.093.5982
Greedy SFTBase Model=Llama 3B2025.093.590.9
Greedy DPOBase Model=Qwen 4B2025.093.560.79
AADBase Model=Qwen 0.6B2025.093.42—
Greedy DPOModel=Llama 1B, Training Protocol=full finetuning2025.093.1563
EFTBase Model=Qwen 0.6B2025.093.080.55
Greedy SFTBase Model=Qwen 4B2025.092.630.88
Bo2Base Model=Qwen 0.6B2025.092.420.68
Greedy SFTModel=Llama 1B, Training Protocol=full finetuning2025.092.0672
Greedy DPOBase Model=Qwen 0.6B2025.091.740.76
Greedy SFTBase Model=Qwen 0.6B2025.090.830.83