List your company
ResearchFormerly WizWand

Question Answering on CommonsenseQA

89.3AccuracyEntropy Equilibrium Sampling (EES)
68.832874.146479.4684.7736May 2, 2020May 10, 2021May 19, 2022May 28, 2023Jun 5, 2024Jun 14, 2025Jun 23, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Entropy Equilibrium Sampling (EES)LM=Qwen2.5-32B, tau=0.52025.1189.3
MirostatLM=Qwen2.5-32B, tau=0.5, Auxiliary Hyperparameter=32025.1189.24
GPT-4oApproach=CoT, Learning Paradigm=Vanilla2024.1186.5
Entropy Equilibrium Sampling (EES)LM=Qwen2.5-7B, tau=0.52025.1184.42
JoBSModel Selection=JoBS, Data Selection=JoBS2026.0284.3
TypicalLM=Qwen2.5-7B, tau=0.8, Auxiliary Hyperparameter=0.22025.1184.11
Entropy Equilibrium Sampling (EES)LM=Qwen2.5-7B, tau=0.82025.1183.95
MirostatLM=Qwen2.5-7B, tau=0.5, Auxiliary Hyperparameter=2.52025.1183.88
Top-pLM=Qwen2.5-7B, tau=0.5, Auxiliary Hyperparameter=0.752025.1183.77
Temperature SamplingLM=Qwen2.5-7B, tau=0.52025.1183.7
Entropy Equilibrium Sampling (EES)LM=Qwen2.5-7B, tau=12025.1183.64
AdaptiveLM=Qwen2.5-7B, tau=0.5, Auxiliary Hyperparameter=0.00052025.1183.62
MirostatLM=Qwen2.5-7B, tau=1, Auxiliary Hyperparameter=2.52025.1183.55
AdaptiveLM=Qwen2.5-7B, tau=0.8, Auxiliary Hyperparameter=0.0052025.1183.46
GPT-4oApproach=Zero-shot, Learning Paradigm=Vanilla2024.1183.4
AdaptiveLM=Qwen2.5-7B, tau=1, Auxiliary Hyperparameter=0.012025.1183.39
TypicalLM=Qwen2.5-7B, tau=0.5, Auxiliary Hyperparameter=0.952025.1183.36
MirostatLM=Qwen2.5-7B, tau=0.8, Auxiliary Hyperparameter=2.52025.1183.23
Qwen 2.5-7BApproach=Ours, Learning Paradigm=Fine-tuned2024.1183.2
EtaLM=Qwen2.5-7B, tau=0.5, Auxiliary Hyperparameter=0.00092025.1182.98
TypicalLM=Qwen2.5-7B, tau=1, Auxiliary Hyperparameter=0.22025.1182.95
Top-kLM=Qwen2.5-7B, tau=0.5, Auxiliary Hyperparameter=102025.1182.75
Top-kLM=Qwen2.5-7B, tau=1, Auxiliary Hyperparameter=202025.1182.65
Top-pLM=Qwen2.5-7B, tau=0.8, Auxiliary Hyperparameter=0.82025.1182.64
Top-kLM=Qwen2.5-7B, tau=0.8, Auxiliary Hyperparameter=1002025.1182.42
Temperature SamplingLM=Qwen2.5-7B, tau=0.82025.1182.39
LLaMA-3.1-8BApproach=Ours, Learning Paradigm=Fine-tuned2024.1182.3
EtaLM=Qwen2.5-7B, tau=1, Auxiliary Hyperparameter=0.00062025.1182.1
Temperature SamplingLM=Qwen2.5-7B, tau=12025.1182.01
Top-pLM=Qwen2.5-7B, tau=1, Auxiliary Hyperparameter=0.82025.1181.93
EtaLM=Qwen2.5-7B, tau=0.8, Auxiliary Hyperparameter=0.0022025.1181.88
MoRAL (layer-wise)Backbone=Mistral-7B, Evaluation Protocol=domain-specific fine-tuning2026.0381.57
LLaMA-3.1-70BApproach=CoT, Learning Paradigm=Vanilla2024.1181.5
AutoLoRAModel Selection=AutoLoRA, Data Selection=BO2026.0281.5
AdaFuseBackbone=Mistral-7B, Evaluation Protocol=domain-specific fine-tuning2026.0381.29
BOModel Selection=BO, Data Selection=IF2026.0281.1
BOModel Selection=BO, Data Selection=BO2026.0280.7
DefaultModel Selection=Default, Data Selection=BO2026.0280.6
PESC (block-wise)Backbone=Mistral-7B, Evaluation Protocol=domain-specific fine-tuning2026.0380.46
RoBoTModel Selection=RoBoT, Data Selection=BO2026.0280.2
Qwen 2.5-7BApproach=Planning-token, Learning Paradigm=Fine-tuned2024.1180.1
DARTSModel Selection=DARTS, Data Selection=Diversity2026.0280.1
LoRABackbone=Mistral-7B, Evaluation Protocol=domain-specific fine-tuning2026.0379.85
AutoLoRAModel Selection=AutoLoRA, Data Selection=LESS2026.0279.8
BOModel Selection=BO, Data Selection=Default2026.0279.7
DARTSModel Selection=DARTS, Data Selection=Default2026.0279.6
DARTSModel Selection=DARTS, Data Selection=BO2026.0279.6
RoBoTModel Selection=RoBoT, Data Selection=IF2026.0279.4
BOModel Selection=BO, Data Selection=LESS2026.0279.4
BOModel Selection=BO, Data Selection=Diversity2026.0279.4
DefaultModel Selection=Default, Data Selection=IF2026.0279.3
FreeLB-ROBERTAseen_dataset_during_training=true2020.0579.1
Qwen 2.5-7BApproach=Zero-shot, Learning Paradigm=Vanilla2024.1178.9
AutoLoRAModel Selection=AutoLoRA, Data Selection=Default2026.0278.9
FULLModel=Qwen3-8b, Zero-shot protocol=true2025.1278.71
Qwen 2.5-7BApproach=CoT, Learning Paradigm=Vanilla2024.1178.4
AGDOTraining Stage=RL2026.0678.3
AGDO-SFTTraining Stage=SFT2026.0678.1
AutoLoRAModel Selection=AutoLoRA, Data Selection=Diversity2026.0278
AutoLoRAModel Selection=AutoLoRA, Data Selection=IF2026.0277.9
Qwen 2.5-7BApproach=LoRA, Learning Paradigm=Fine-tuned2024.1177.5
DefaultModel Selection=Default, Data Selection=Diversity2026.0277.4
FULLModel=Llama3-8b, Zero-shot protocol=true2025.1277.23
RoBoTModel Selection=RoBoT, Data Selection=DoReMi2026.0277.1
Entropy Equilibrium Sampling (EES)LM=Llama3.1-8B, tau=Avg2025.1177.02
BOModel Selection=BO, Data Selection=DoReMi2026.0277
LLaMA-3.1-8BApproach=Planning-token, Learning Paradigm=Fine-tuned2024.1176.9
LLaMA-3.1-70BApproach=Zero-shot, Learning Paradigm=Vanilla2024.1176.5
DefaultModel Selection=Default, Data Selection=Default2026.0276.3
DARTSModel Selection=DARTS, Data Selection=LESS2026.0276.3
RoBoTModel Selection=RoBoT, Data Selection=Diversity2026.0276.3
UnifiedQAseen_dataset_during_training=false, training_format=multi-format2020.0576.2
DARTSModel Selection=DARTS, Data Selection=DoReMi2026.0276.1
AutoLoRAModel Selection=AutoLoRA, Data Selection=DoReMi2026.0276.1
UnifiedQA [MC]seen_dataset_during_training=false, training_format=multiple choice2020.0575.9
RoBoTModel Selection=RoBoT, Data Selection=LESS2026.0275.5
SFTTraining Stage=SFT2026.0675.5
LLaMA-3.1-8BApproach=LoRA, Learning Paradigm=Fine-tuned2024.1175.4
Temperature SamplingLM=Llama3.1-8B, tau=Avg2025.1175.34
KALEBackbone=Qwen2.5 32B, Category=Augmented-based2026.0175.02
ExpGraphBackbone Model=Llama-3.1-8B-Instruct (Large LLM), Method Category=LLM-Centric Experience Learning Baselines2026.0575
RoBoTModel Selection=RoBoT, Data Selection=Default2026.0274.9
TEALModel=Qwen3-8b, Zero-shot protocol=true2025.1274.77
SPONModel=Qwen3-8b, Zero-shot protocol=true2025.1274.28
DefaultModel Selection=Default, Data Selection=DoReMi2026.0274.2
DARTSModel Selection=DARTS, Data Selection=IF2026.0273.7
DefaultModel Selection=Default, Data Selection=LESS2026.0273
Blockwise SFTTraining Stage=SFT2026.0672.9
Roberta-largeEvaluation Protocol=fine-tuned, Backbone=Roberta-large2026.0672.1
MTO-MAPFull name=Masked Answer Prompting, Backbone=T5-large2026.0671.74
BaseModel=LLaMA-3.1-8B, Compression=0%, Mode=Zero-shot2025.1071.66
StructGPTBackbone=Qwen2.5 32B, Category=Retrieval-based2026.0171.58
MTO-D+MCPFull name=Masked Choice Prompting with a Denoising objective, Backbone=T5-large2026.0671.58
SDFTBackbone=Qwen2.5 32B, Category=SFT-based2026.0171.25
T5-largeEvaluation Protocol=fine-tuned, Backbone=T5-large, Settings=using our settings2026.0671.25
LLaMA-2-7BApproach=Ours, Learning Paradigm=Fine-tuned2024.1171.1
LLaMA-3.1-8BApproach=CoT, Learning Paradigm=Vanilla2024.1170.6
DMTBackbone=Qwen2.5 32B, Category=SFT-based2026.0170.43
BaselineBit width (b)=16.002025.0570.2
KG-SFTBackbone=Qwen2.5 32B, Category=SFT-based2026.0169.62