Commonsense Reasoning on HellaSwag (val)

95.3Accuracy

GPT-4

Updated 5d ago

Evaluation Results

Method	Links
GPT-4 2023.03		95.3
D2-LoRA 2026.02		91
LoRA 2026.02		90.2
DoRA 2026.02		90
ALUM 2023.03		85.6
GPT-3.5 2023.03		85.5
LLaMA 2023.03		84.2
D2-LoRA 2026.02		82.4
DoRA 2026.02		81.8
LoRA 2026.02		77.6
EvoGM 2026.05		66
PSO-Merging 2026.05		65
CMA 2026.05		64.5
EvoGM 2026.05		63.5
Model Swarm 2026.05		63.5
Single Best 2026.05		62
Single Best 2026.05		62
DARE 2026.05		61.5
DARE 2026.05		61.5
Task Arithmetic 2026.05		60.5
TA 2026.05		60.5
CMA 2026.05		60
Base 2026.05		59
Model Soup 2026.05		59
Base 2026.05		59
Model Soup 2026.05		59
Model Swarm 2026.05		58.5
DeepSeekMoE 2024.01		54.8
PSO-Merging 2026.05		53
MTL 2026.05		52.5
MTL 2026.05		52.5
TIES 2026.05		51
TIES 2026.05		51
Mistral (Full-Attention) 2024.07		50.7
GShard 2024.01		50.5
Switch 2024.01		49.1
BMoJo (Fading + Eidetic) 2024.07		46.8
Hash Layer 2024.01		46.2
BMoJo (Fading) 2024.07		46
Mamba (SSM) 2024.07		45
Dense 2024.01		38.8
Hybrid (Sliding Attention + SSM) 2024.07		38.8
AMO 2026.05		37.46
Mistral (Full-Attention) 2024.07		33.8
AMO 2026.05		33.79
Mamba (SSM) 2024.07		33.4
BMoJo (Fading + Eidetic) 2024.07		33.3
BMoJo (Fading) 2024.07		33.2
TMMFormer 2026.05		31.82
YuriiFormer 2026.05		31.58
Hybrid (Sliding Attention + SSM) 2024.07		31.3
AdamFormer 2026.05		30.96
VanillaTransformer 2026.05		30.2
AdamWFormer 2026.05		30.08