List your company

Assistant Response Alignment (Helpfulness and Harmlessness) on HH-RLHF (test)

89.42Helpfulness Win RateMetaAligner-7B
-87.8168-41.80344.2150.2234Oct 25, 2023Mar 2, 2024Jul 10, 2024Nov 17, 2024Mar 26, 2025Aug 3, 2025Dec 11, 2025

Evaluation results

MethodLinks
MetaAligner-7BPolicy Model=LLaMA2-Chat-70B, GPU Hours=242.682024.0389.42—85.1688.0887.55——————
MetaAligner-13BPolicy Model=LLaMA2-Chat-7B, GPU Hours=403.442024.0385.5—76.58682.67——————
Self-RefinementPolicy Model=LLaMA2-Chat-70B2024.0382.8—70.4868.9174.06——————
MetaAligner-7BPolicy Model=LLaMA2-Chat-7B, GPU Hours=242.682024.0382—77.58380.83——————
MetaAligner-7B (w/o Equal Pref.)Policy Model=LLaMA2-Chat-7B, GPU Hours=242.682024.0380.7—73.8277.3977.3——————
MetaAligner-7B (w/o Warm Up)Policy Model=LLaMA2-Chat-7B, GPU Hours=242.682024.0380.32—77.182.6380.02——————
RLHF-r1Alignment Algorithm=RLHF, Objective=r1, Base Model=phi-1.52025.1276————————-0.42-0.23
SFTPolicy Model=LLaMA2-Chat-7B, GPU Hours=247.742024.0375—7276.574.5——————
Aligner-7BPolicy Model=LLaMA2-Chat-7B, GPU Hours=236.82024.0375—66.570.1270.54——————
MetaAligner-1.1BPolicy Model=LLaMA2-Chat-7B, GPU Hours=120.482024.0375—62.57771.5——————
ChatGPTprotocol=zero-shot2023.1068.28—72.19—70.4369.8571.02————
CycleAlign_PROBackbone=Alpaca2023.1067.89—71.32—68.9766.5368.92————
CycleAlign_RRHFBackbone=LLaMA2023.1067.05—71.66—68.4365.8967.95————
CycleAlign_PROBackbone=LLaMA2023.1066.49—70.62—68.4167.6768.5————
PROBackbone=Alpaca2023.1064.56—73.13—67.6465.666.51————
PROBackbone=LLaMA2023.1064.05—72.86—67.465.5666.44————
MODPOPolicy Model=LLaMA2-Chat-7B, GPU Hours=405.92024.0364—657869——————
ChatGLMprotocol=zero-shot, Backbone=ChatGLM-6B2023.1062.14—67.26—63.8560.4463.86————
RRHFBackbone=LLaMA2023.1061.38—64.63—63.1263.2663.28————
PPOBackbone=LLaMA2023.1055.29—61.97—58.6559.7858.26————
Alpacaprotocol=zero-shot, Backbone=Alpaca-7B2023.1053.85—52.77—54.2655.355.43————
MORLHFPolicy Model=LLaMA2-Chat-7B, GPU Hours=1892.32024.0351.2—62.8377.563.84——————
MOPO-LagAlignment Algorithm=MOPO, Variant=Lagrangian (Lag), Base Model=phi-1.52025.1239————————0.220.17
LLaMAprotocol=zero-shot, Backbone=LLaMA-7B2023.1033.25—53.59—40.6740.4836.23————
PARMAlignment Algorithm=PARM, Base Model=phi-1.52025.1231————————0.170.23
MOPO-LBAlignment Algorithm=MOPO, Variant=Lower Bound (LB), Base Model=phi-1.52025.1230————————0.190.18
RiCAlignment Algorithm=RiC, Base Model=phi-1.52025.1225————————0.150.11
DPOAlignment Algorithm=DPO, Alignment Dataset=D_J, Base Model=phi-1.52025.1218————————0.090.11
MODPOAlignment Algorithm=MODPO, Base Model=phi-1.52025.124————————-0.090.08
RLHF-r3Alignment Algorithm=RLHF, Objective=r3, Base Model=phi-1.52025.12-79————————-0.920.42
RLHF-r2Alignment Algorithm=RLHF, Objective=r2, Base Model=phi-1.52025.12-81————————0.53-0.4
DLMA-7BBase Model=Mistral-7B2024.02————43——1740——
DLMA-7BBase Model=Falcon-7B2024.02————44——1541——
MAT-STEERcomparison_baseline=ICL2025.02—67.07—————————
MAT-STEERcomparison_baseline=SFT2025.02—62.35—————————
MAT-STEERcomparison_baseline=LITO2025.02—59.12—————————