List your company

Zero-shot Commonsense Reasoning on ARC-e, ARC-c, HellaSwag, PIQA, WinoGrande (test)

82.87ARC-e AccuracyFP16
34.229246.857159.48572.1129Jan 28, 2026Feb 14, 2026Mar 3, 2026Mar 21, 2026Apr 7, 2026Apr 24, 2026May 12, 2026

Evaluation results

MethodLinks
FP16Model=LLaMA-3.1 8B-Instruct, Precision=FP162026.0582.8754.7879.381.0174.4374.48
RaZeRModel=LLaMA-3.1 8B-Instruct, Quantization=W4A162026.0581.6954.9678.8581.0673.2473.96
4over6Model=LLaMA-3.1 8B-Instruct, Quantization=W4A162026.0581.5755.0378.9481.0773.2473.97
SOARModel=Qwen3 8B, Quantization=W4A162026.0581.4456.9973.9677.869.6971.98
FP16Model=Qwen3 8B, Precision=FP162026.0580.8956.6674.8677.5867.6471.53
NVFP4Model=LLaMA-3.1 8B-Instruct, Quantization=W4A162026.0580.3953.578.8480.6373.2473.32
NVFP4Model=Qwen3 8B, Quantization=W4A162026.0580.2256.6674.0676.7168.4371.22
SOARModel=LLaMA-3.1 8B-Instruct, Quantization=W4A162026.0580.1853.927980.273.873.42
RaZeRModel=Qwen3 8B, Quantization=W4A162026.0579.9756.4874.2477.268.0371.18
4over6Model=Qwen3 8B, Quantization=W4A162026.0579.6355.873.9776.1768.1970.75
FP16Model=Qwen3 4B, Precision=FP162026.0578.4554.0168.4675.0865.968.38
SOARModel=Qwen3 4B, Quantization=W4A162026.0576.7352.1367.8875.0363.6167.08
NVFP4Model=Qwen3 4B, Quantization=W4A162026.0576.4750.6867.2674.762.9866.42
4over6Model=Qwen3 4B, Quantization=W4A162026.0576.2651.6267.2674.7664.466.86
RaZeRModel=Qwen3 4B, Quantization=W4A162026.0575.9351.4567.5475.1963.7766.78
BF16Backbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0174.542.255.276.869.163.6
HESTIABackbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0174.139.349.175.762.260.1
SOARModel=LLaMA-3.2 3B-Instruct, Quantization=W4A162026.0571.4245.7371.376.2268.7566.68
FP16Model=LLaMA-3.2 3B-Instruct, Precision=FP162026.0571.0946.2571.5576.6668.5966.83
NVFP4Model=LLaMA-3.2 3B-Instruct, Quantization=W4A162026.0570.6245.1371.2476.567.6466.23
4over6Model=LLaMA-3.2 3B-Instruct, Quantization=W4A162026.0570.5444.870.8476.568.1966.17
RaZeRModel=LLaMA-3.2 3B-Instruct, Quantization=W4A162026.0570.4545.9971.2576.568.5166.54
TequilaBackbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0170.234.646.473.962.757.6
TWNBackbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0169.235.146.273.458.656.5
INT2Backbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0168.734.842.372.755.854.9
AbsMeanBackbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0167.232.943.973.558.255.1
HESTIABackbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0166.53541.872.657.754.7
BF16Backbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0165.431.347.774.260.355.8
TequilaBackbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0164.530.539.17154.251.9
FP16Model=LLaMA-3.2 1B-Instruct, Precision=FP162026.0563.7237.5461.5974.9261.6459.88
SOARModel=LLaMA-3.2 1B-Instruct, Quantization=W4A162026.0562.2937.6359.4173.7861.6458.95
INT2Backbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0161.926.235.469.252.749.1
RaZeRModel=LLaMA-3.2 1B-Instruct, Quantization=W4A162026.0561.737.5460.0273.8362.1959.06
NVFP4Model=LLaMA-3.2 1B-Instruct, Quantization=W4A162026.0561.4137.1259.2273.5662.6758.8
4over6Model=LLaMA-3.2 1B-Instruct, Quantization=W4A162026.0561.3636.659.5273.1861.4858.43
AbsMeanBackbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0160.325.93668.354.148.9
SEQBackbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0149.823.130.364.552.944.1
LSQBackbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0143.12029.459.952.240.9
DLTBackbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0142.417.425.656.351.338.6
SEQBackbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0142.11827.360.45139.8
TWNBackbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0140.72228.460.149.240.1
LSQBackbone=LLaMA-3.2-1B, Evaluation Protocol=Zero-shot2026.0137.617.725.857.450.637.8
DLTBackbone=LLaMA-3.2-3B, Evaluation Protocol=Zero-shot2026.0136.116.12657.249.637