List your company
ResearchFormerly WizWand

Commonsense Reasoning on HellaS (Accuracy)

93.49AccuracyABBA
39.316453.380767.44581.5093Sep 25, 2025Sep 26, 2025
Updated 5mo ago

Evaluation results

MethodLinks
ABBAModel=Llama-3.2-3B, #Params=48.63M2025.0993.49
BoHAModel=Llama-3.2-3B, #Params=48.63M2025.0993.16
FFTModel=Llama-3.2-3B, #Params=3.21B2025.0993.14
HiRAModel=Llama-3.2-3B, #Params=48.63M2025.0992.94
GraLoRAModel=Llama-3.2-3B, #Params=48.63M2025.0992.38
rsLoRAModel=Llama-3.2-3B, #Params=48.63M2025.0990.45
DoRAModel=Llama-3.2-3B, #Params=49.40M2025.0990.43
LoRAModel=Llama-3.2-3B, #Params=48.63M2025.0989.63
PiSSAModel=Llama-3.2-3B, #Params=48.63M2025.0988.05
EMoEBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=12 + 22025.0984.54
EMoEBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=6 + 22025.0984.18
EMoEBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=3 + 22025.0982.96
Top-kBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=6 + 22025.0981.56
ABBAModel=Llama-3.2-1B, #Params=22.54M2025.0980.99
Top-kBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=3 + 22025.0980.86
Top-kBase Model=ERNIE-4.5-21B-A3B, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=12 + 22025.0980.08
BoHAModel=Llama-3.2-1B, #Params=22.54M2025.0979.82
FFTModel=Llama-3.2-1B, #Params=1.24B2025.0979.63
GraLoRAModel=Llama-3.2-1B, #Params=22.54M2025.0977.81
DoRAModel=Llama-3.2-1B, #Params=22.92M2025.0977.48
LoRAModel=Llama-3.2-1B, #Params=22.54M2025.0977.41
PiSSAModel=Llama-3.2-1B, #Params=22.54M2025.0976.82
HiRAModel=Llama-3.2-1B, #Params=22.54M2025.0975.74
rsLoRAModel=Llama-3.2-1B, #Params=22.54M2025.0975.34
EMoEBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=12 + 22025.0962.25
EMoEBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=6 + 22025.0961.29
EMoEBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=EMoE, Inference Budget (k')=3 + 22025.0959.61
Top-kBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=6 + 22025.0958.52
Top-kBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=12 + 22025.0957.6
Top-kBase Model=DeepSeek-V2-Lite, Trained Activated Experts=6+2, MoE Routing Method=Top-k, Inference Budget (k')=3 + 22025.0957.36
EMoEBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=EMoE, Inference Budget (k')=62025.0956.14
EMoEBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=EMoE, Inference Budget (k')=42025.0955.95
Top-kBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=Top-k, Inference Budget (k')=22025.0954.87
EMoEBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=EMoE, Inference Budget (k')=22025.0953.57
Top-kBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=Top-k, Inference Budget (k')=42025.0953.07
Top-kBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=Top-k, Inference Budget (k')=12025.0952.71
EMoEBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=EMoE, Inference Budget (k')=12025.0951.52
EMoEBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=EMoE, Inference Budget (k')=82025.0951.22
Top-kBase Model=LoRAMoE, Trained Activated Experts=2, MoE Routing Method=Top-k, Inference Budget (k')=62025.0950.76
EMoEBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=EMoE, Inference Budget (k')=162025.0950.69
Top-kBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=Top-k, Inference Budget (k')=82025.0948.68
Top-kBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=Top-k, Inference Budget (k')=162025.0948.59
EMoEBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=EMoE, Inference Budget (k')=42025.0943.24
Top-kBase Model=OLMoE-1B-7B-0924, Trained Activated Experts=8, MoE Routing Method=Top-k, Inference Budget (k')=42025.0941.4