List your company
ResearchFormerly WizWand

LLM Alignment on AlpacaEval

25.24Win RateHighAvg.
3.09848.846714.59520.3433May 29, 2025
Updated 5mo ago

Evaluation results

MethodLinks
HighAvg.Backbone=LLaMA-3-8B, % Train=33%, Alignment Method=SimPO2025.0525.24——
FullBackbone=LLaMA-3-8B, % Train=100%, Alignment Method=DPO2025.0518.19——
HighAvg.Backbone=LLaMA-3-8B, % Train=33%, Alignment Method=DPO2025.0517.73——
FullBackbone=LLaMA-3-8B, % Train=100%, Alignment Method=SimPO2025.0517.54——
RandomBackbone=LLaMA-3-8B, % Train=33%, Alignment Method=DPO2025.0513.47——
LowAvg.Backbone=LLaMA-3-8B, % Train=33%, Alignment Method=DPO2025.0511.55——
HighVar.Backbone=LLaMA-3-8B, % Train=33%, Alignment Method=DPO2025.058.59——
LowAvg.Backbone=Mistral-7B, % Train=33%, Alignment Method=DPO2025.057.2——
HighVar.Backbone=Mistral-7B, % Train=33%, Alignment Method=DPO2025.056.86——
RandomBackbone=Mistral-7B, % Train=33%, Alignment Method=DPO2025.056.82——
FullBackbone=Mistral-7B, % Train=100%, Alignment Method=DPO2025.056.81——
HighAvg.Backbone=Mistral-7B, % Train=33%, Alignment Method=DPO2025.056.65——
HighAvg.Backbone=Mistral-7B, % Train=33%, Alignment Method=SimPO2025.055.79——
FullBackbone=Mistral-7B, % Train=100%, Alignment Method=SimPO2025.055.32——
LowAvg.Backbone=Mistral-7B, % Train=33%, Alignment Method=SimPO2025.055.2——
LowAvg.Backbone=LLaMA-3-8B, % Train=33%, Alignment Method=SimPO2025.055.17——
HighVar.Backbone=Mistral-7B, % Train=33%, Alignment Method=SimPO2025.055.03——
RandomBackbone=LLaMA-3-8B, % Train=33%, Alignment Method=SimPO2025.054.9——
RandomBackbone=Mistral-7B, % Train=33%, Alignment Method=SimPO2025.054.77——
ZeroshotBackbone=Mistral-7B, % Train=0%, Alignment Method=DPO2025.054.3——
ZeroshotBackbone=Mistral-7B, % Train=0%, Alignment Method=SimPO2025.054.3——
HighVar.Backbone=LLaMA-3-8B, % Train=33%, Alignment Method=SimPO2025.054.23——
ZeroshotBackbone=LLaMA-3-8B, % Train=0%, Alignment Method=DPO2025.053.95——
ZeroshotBackbone=LLaMA-3-8B, % Train=0%, Alignment Method=SimPO2025.053.95——
AdaBoNLM=Mistral, RM=Mistral, K=5, B=120, d=0.75B2025.05—94—
AdaBoNLM=Mistral, RM=FsfairX, K=5, B=120, d=0.75B2025.05—92—
AdaBoNLM=Mistral, RM=Armo, K=5, B=120, d=0.75B2025.05—96—
AdaBoNLM=Qwen, RM=Mistral, K=5, B=120, d=0.75B2025.05—100—
AdaBoNLM=Qwen, RM=FsfairX, K=5, B=120, d=0.75B2025.05—98—
AdaBoNLM=Qwen, RM=Armo, K=5, B=120, d=0.75B2025.05—78—
AdaBoNLM=Gemma, RM=Mistral, K=5, B=120, d=0.75B2025.05—76—
AdaBoNLM=Gemma, RM=FsfairX, K=5, B=120, d=0.75B2025.05—92—
AdaBoNLM=Gemma, RM=Armo, K=5, B=120, d=0.75B2025.05—86—
AdaBoNLM=Llama, RM=Mistral, K=5, B=120, d=0.75B2025.05—92—
AdaBoNLM=Llama, RM=FsfairX, K=5, B=120, d=0.75B2025.05—96—
AdaBoNLM=Llama, RM=Armo, K=5, B=120, d=0.75B2025.05—92—
Mistral-7B-INSTFine-tuning Data=HelpSteer2 prompts2024.08——10.31
Qwen2-1.5B-INSTFine-tuning Data=HelpSteer2 prompts2024.08——1.06
RLHFBase Model=Qwen2-1.5B-INST, Fine-tuning Data=HelpSteer2 prompts2024.08——0.66
RLHFBase Model=Mistral-7B-INST, Fine-tuning Data=HelpSteer2 prompts2024.08——10.15
UNABase Model=Qwen2-1.5B-INST, Fine-tuning Data=HelpSteer2 prompts2024.08——1.63
UNABase Model=Mistral-7B-INST, Fine-tuning Data=HelpSteer2 prompts2024.08——10.54