List your company
ResearchFormerly WizWand

Commonsense Reasoning on HellaSwag (HellaSwag Score)

86HellaSwag ScoreTILEQ_s
27.03242.34157.6572.959Nov 14, 2025Dec 16, 2025Jan 18, 2026Feb 20, 2026Mar 24, 2026Apr 26, 2026May 29, 2026
Updated 5mo ago

Evaluation results

MethodLinks
TILEQ_sModel=Mixtral-8x7B, Avg.bit=3.162026.0586
DEITABase Model=Llama-3-8B, Training Data Scale=15%2026.0584.45
MoDSBase Model=Llama-3-8B, Training Data Scale=15%2026.0584.43
InsTagBase Model=Llama-3-8B, Training Data Scale=15%2026.0584.28
MADS3BBase Model=Llama-3-8B, Training Data Scale=15%, Instructor Model=Llama-3.2-3B-Instruct2026.0584.24
FullBase Model=Llama-3-8B, Training Data Scale=100%2026.0584.16
NUGGETSBase Model=Llama-3-8B, Training Data Scale=15%2026.0584.11
MADS8BBase Model=Llama-3-8B, Training Data Scale=15%, Instructor Model=Llama-3.1-8B-Instruct2026.0584.07
SelectITBase Model=Llama-3-8B, Training Data Scale=15%2026.0584.02
ClusterClipBase Model=Llama-3-8B, Training Data Scale=15%2026.0583.94
IFDBase Model=Llama-3-8B, Training Data Scale=15%2026.0583.38
LLAMA-4-SCOUTPARAMS=109B, Variant=ORIGINAL2025.1182.9
MiLoModel=Mixtral-8x7B, Avg.bit=3.42026.0582.2
LLAMA-4-SCOUTPARAMS=109B, Variant=FCSD2025.1182
DEITABase Model=Llama-2-7B, Training Data Scale=15%2026.0581.67
SelectITBase Model=Llama-2-7B, Training Data Scale=15%2026.0581.34
InsTagBase Model=Llama-2-7B, Training Data Scale=15%2026.0581.33
NUGGETSBase Model=Llama-2-7B, Training Data Scale=15%2026.0581.32
MADS3BBase Model=Llama-2-7B, Training Data Scale=15%, Instructor Model=Llama-3.2-3B-Instruct2026.0581.31
MADS8BBase Model=Llama-2-7B, Training Data Scale=15%, Instructor Model=Llama-3.1-8B-Instruct2026.0581.26
FullBase Model=Llama-2-7B, Training Data Scale=100%2026.0581.25
MoDSBase Model=Llama-2-7B, Training Data Scale=15%2026.0581.04
ClusterClipBase Model=Llama-2-7B, Training Data Scale=15%2026.0581.02
IFDBase Model=Llama-2-7B, Training Data Scale=15%2026.0580.82
DEEPSEEK-V2-LITEPARAMS=16B, Variant=ORIGINAL2025.1180.8
LLAMA-4-SCOUTPARAMS=109B, Variant=SFT2025.1180
QWEN-3-30B MOEPARAMS=30B, Variant=ORIGINAL2025.1179.7
Qwen2.5-7BParameters=7B2026.0578.9
TILEQ_sModel=Deepseek-MoE, Avg.bit=3.162026.0578.5
FP16Bit-width=FP16, Evaluation Protocol=zero-shot2026.0578.26
QWEN-3-30B MOEPARAMS=30B, Variant=FCSD2025.1177.2
DEEPSEEK-V2-LITEPARAMS=16B, Variant=FCSD2025.1176.9
QWEN-3-30B MOEPARAMS=30B, Variant=SFT2025.1175.6
Qwen3.5-4BParameters=4B2026.0575.3
MiLoModel=Deepseek-MoE, Avg.bit=3.42026.0574.6
OLMo-3-7BParameters=7B2026.0574.2
DEEPSEEK-V2-LITEPARAMS=16B, Variant=SFT2025.1174.1
Mellum 2Parameters=2.5B/12B2026.0573.7
Qwen3-4BParameters=4B2026.0573.7
QTIP 2-bit PTQBit-width=2-bit, Evaluation Protocol=zero-shot2026.0571.15
BCJR-QAT-N4Bit-width=2 bpw, Evaluation Protocol=zero-shot, Configuration=N42026.0570.82
I-DPOBase Model=Qwen2.5-7B-Instruct2026.0562.2
I-DPO + MaPPOBase Model=Qwen2.5-7B-Instruct2026.0562.2
ITBase Model=Qwen2.5-7B-Instruct2026.0562.1
DPO + MaPPOBase Model=Qwen2.5-7B-Instruct2026.0562.1
DPOBase Model=Qwen2.5-7B-Instruct2026.0562
Echo-180MParams=180M, Tokens=10B, Zero-shot=true2026.0544.3
Mamba-3-MIMO-180MParams=180M, Tokens=100B, Zero-shot=true2026.0541
Mamba-3-SISO-180MParams=180M, Tokens=100B, Zero-shot=true2026.0540.8
GDN-180MParams=180M, Tokens=100B, Zero-shot=true2026.0540.2
Mamba-2-180MParams=180M, Tokens=100B, Zero-shot=true2026.0540.1
Transformer-180MParams=180M, Tokens=100B, Zero-shot=true2026.0539
Echo-50MParams=50M, Tokens=3B, Zero-shot=true2026.0529.3