List your company

Alignment on UltraFeedback (test)

63.72Honesty ScoreMetaAligner-7B
25.531235.445645.3655.2744Mar 25, 2024Jun 25, 2024Sep 26, 2024Dec 27, 2024Mar 30, 2025Jun 30, 2025Oct 1, 2025

Evaluation results

MethodLinks
MetaAligner-7BPolicy Model=LLaMA2-Chat-70B, GPU Hours=242.682024.0363.7267.0570.154.763.89
Self-RefinementPolicy Model=LLaMA2-Chat-70B2024.0362.9149.9560.7757.655.05
MetaAligner-13BPolicy Model=LLaMA2-Chat-7B, GPU Hours=403.442024.0359.568.5645561.75
MetaAligner-7BPolicy Model=LLaMA2-Chat-7B, GPU Hours=242.682024.034851.3355.6744.3349.83
MetaAligner-7B (w/o Warm Up)Policy Model=LLaMA2-Chat-7B, GPU Hours=242.682024.0347.449.9655.7344.1849.32
MetaAligner-7B (w/o Equal Pref.)Policy Model=LLaMA2-Chat-7B, GPU Hours=242.682024.0343.646.853.1741.746.32
MODPOPolicy Model=LLaMA2-Chat-7B, GPU Hours=405.92024.0343.430.8237.192534.1
MAH-DPOBackbone=Llama-3.1-8B-Instruct, Configuration=Ensemble2025.1036.87—24.7863.89—
SFTPolicy Model=LLaMA2-Chat-7B, GPU Hours=247.742024.0336.5272636.531.5
Aligner-7BPolicy Model=LLaMA2-Chat-7B, GPU Hours=236.82024.0336.5272636.531.5
MAH-DPOBackbone=Llama-3.1-8B-Instruct, Configuration=Honest Head2025.1035.16—23.0362.57—
MODPOBackbone=Llama-3.1-8B-Instruct2025.1034.77—23.2561.75—
MAH-DPOBackbone=Llama-3.1-8B-Instruct, Configuration=Help Head2025.1034.65—22.3963.09—
MAH-DPOBackbone=Llama-3.1-8B-Instruct, Configuration=Truth Head2025.1034.61—22.8662.57—
MORLHFPolicy Model=LLaMA2-Chat-7B, GPU Hours=1892.32024.0333.732.1826.133.731.42
DPO SoupBackbone=Llama-3.1-8B-Instruct2025.1032.17—21.5361.28—
Single-Head DPOBackbone=Llama-3.1-8B-Instruct2025.1030.55—20.1460.43—
BaseBackbone=Llama-3.1-8B-Instruct2025.1030.42—18.8858—
SFTBackbone=Llama-3.1-8B-Instruct2025.1029.98—19.9255.46—
MetaAligner-1.1BPolicy Model=LLaMA2-Chat-7B, GPU Hours=120.482024.032727.673325.3328.25