List your company

Commonsense Reasoning on HellaSwag (HellaSwag Score and AVG)

86.86HellaSwag ScoreUNA-score & binary
49.981659.555869.1378.7042Aug 27, 2024Dec 5, 2024Mar 16, 2025Jun 25, 2025Oct 3, 2025Jan 12, 2026Apr 23, 2026

Evaluation results

MethodLinks
UNA-score & binaryBase Model=Mistral 7B, Alignment Strategy=UNA-score & binary2024.0886.86—
UNA-score & binaryBase Model=Llama 8B, Alignment Strategy=UNA-score & binary2024.0885.4—
UNA-binary (BCE)Base Model=Mistral 7B, Alignment Strategy=UNA-binary (BCE)2024.0884.6—
UNA-score (MSE)Base Model=Mistral 7B, Alignment Strategy=UNA-score (MSE)2024.0884.52—
DPOBase Model=Mistral 7B, Alignment Strategy=DPO2024.0884.42—
UNA-pairwiseBase Model=Mistral 7B, Alignment Strategy=UNA-pairwise2024.0884.42—
KTOBase Model=Mistral 7B, Alignment Strategy=KTO2024.0884.21—
UNA-score (MSE)Base Model=Llama 8B, Alignment Strategy=UNA-score (MSE)2024.0884.14—
Mistral 7BBase Model=Mistral 7B, Alignment Strategy=Baseline2024.0883.44—
KTOBase Model=Llama 8B, Alignment Strategy=KTO2024.0883.15—
DPOBase Model=Llama 8B, Alignment Strategy=DPO2024.0883.01—
UNA-pairwiseBase Model=Llama 8B, Alignment Strategy=UNA-pairwise2024.0883.01—
UNA-binary (BCE)Base Model=Llama 8B, Alignment Strategy=UNA-binary (BCE)2024.0883—
UNA-score & binaryBase Model=Qwen 8B, Alignment Strategy=UNA-score & binary2024.0882.92—
Llama 8BBase Model=Llama 8B, Alignment Strategy=Baseline2024.0881.78—
UNA-score & binaryBase Model=Gemma 4B, Alignment Strategy=UNA-score & binary2024.0881.21—
UNA-score (MSE)Base Model=Qwen 8B, Alignment Strategy=UNA-score (MSE)2024.0881—
HT-MNPOReward Model=Athene-RM-8B2025.0980.4470.99
DPO2025.0980.3770.68
SFT Model2025.0980.370.28
HT-MNPOReward Model=ArmoRM-Llama32025.0980.2970.83
INPO2025.0980.2270.25
HT-MNPOReward Model=Skywork-Reward-V22025.0980.2271.8
TD-MNPO2025.0980.1871.08
SPPO2025.0980.170.19
Moderate-GiModel=Qwen2.52026.0479.9—
DPOBase Model=Qwen 8B, Alignment Strategy=DPO2024.0879.81—
UNA-pairwiseBase Model=Qwen 8B, Alignment Strategy=UNA-pairwise2024.0879.81—
UNA-score (MSE)Base Model=Gemma 4B, Alignment Strategy=UNA-score (MSE)2024.0879.71—
Qwen 8BBase Model=Qwen 8B, Alignment Strategy=Baseline2024.0879.71—
KTOBase Model=Qwen 8B, Alignment Strategy=KTO2024.0879.58—
RandomModel=Qwen2.52026.0479.4—
UNA-binary (BCE)Base Model=Qwen 8B, Alignment Strategy=UNA-binary (BCE)2024.0879.34—
O-LoRAModel=Qwen2.52026.0479.3—
Grad. ClipModel=Qwen2.52026.0479.3—
BaselineModel=Qwen2.52026.0479.2—
EWCModel=Qwen2.52026.0479.2—
KLModel=Qwen2.52026.0479.1—
SimPO2025.0978.969.6
O-LoRAModel=LLaMA-3.12026.0478.3—
RandomModel=LLaMA-3.12026.0478—
KLModel=LLaMA-3.12026.0478—
EWCModel=LLaMA-3.12026.0478—
UNA-binary (BCE)Base Model=Gemma 4B, Alignment Strategy=UNA-binary (BCE)2024.0877.95—
BaselineModel=LLaMA-3.12026.0477.9—
Moderate-GiModel=LLaMA-3.12026.0477.9—
DPOBase Model=Gemma 4B, Alignment Strategy=DPO2024.0877.83—
UNA-pairwiseBase Model=Gemma 4B, Alignment Strategy=UNA-pairwise2024.0877.83—
KTOBase Model=Gemma 4B, Alignment Strategy=KTO2024.0877.56—
Gemma 4BBase Model=Gemma 4B, Alignment Strategy=Baseline2024.0877.48—
DPModel Scale=Gemma 9B, Method=DP2026.0472.1—
Decoupled DiLoCoModel Scale=Gemma 9B, Method=DiLoCo2026.0472—
EWCModel=Qwen32026.0471.1—
BaselineModel=Qwen32026.0471—
KLModel=Qwen32026.0471—
O-LoRAModel=Qwen32026.0470.7—
RandomModel=Qwen32026.0470.6—
Moderate-GiModel=Qwen32026.0470.1—
DPModel Scale=Gemma 5B, Method=DP2026.0465.8—
Decoupled DiLoCoModel Scale=Gemma 5B, Method=DiLoCo2026.0464.9—
DPModel Scale=Gemma 2B, Method=DP2026.0452.1—
Decoupled DiLoCoModel Scale=Gemma 2B, Method=DiLoCo2026.0451.4—