List your company

Mathematical Reasoning on GSM8k (Accuracy, Delta)

78.09AccuracyDense
-0.253220.085940.42560.7641Mar 17, 2026

Evaluation results

MethodLinks
DenseModel=Llama-3.1-8B-Instruct2026.0378.09—
DenseModel=Llama-3.1-8B-Instruct, Compression Method=Dense2026.0378.09—
DenseModel=Llama-3.1-8B-Instruct2026.0378.09—
ZipCalModel=Llama-3.1-8B-Instruct, Calibration Category=Mean2026.0376.250.43
COLAModel=Llama-3.1-8B-Instruct, Calibration Category=Mean2026.0375.82—
ZipCalModel=Llama-3.1-8B-Instruct, Compression Method=Wanda, Compression Setting (Sparsity/Bit-width)=25%2026.0375.8-0.02
DenseModel=gemma-2-9b-it, Compression Method=Dense2026.0375.44—
DenseModel=gemma-2-9b-it2026.0375.44—
ZipCalModel=gemma-2-9b-it, Compression Method=Wanda, Compression Setting (Sparsity/Bit-width)=25%2026.0374.750.44
ZipCalModel=gemma-2-9b-it, Calibration Category=Mean2026.0374.490.18
COLAModel=gemma-2-9b-it, Calibration Category=Mean2026.0374.31—
ZipCalModel=gemma-2-9b-it, Compression Method=AWQ, Compression Setting (Sparsity/Bit-width)=W4A162026.0374.1-0.12
ZipCalModel=Llama-3.1-8B-Instruct, Compression Method=GPTQ, Compression Setting (Sparsity/Bit-width)=W4A162026.0374.076.46
ZipCalModel=gemma-2-9b-it, Compression Method=GPTQ, Compression Setting (Sparsity/Bit-width)=W4A162026.0373.61-0.55
ZipCalModel=Llama-3.1-8B-Instruct, Compression Method=AWQ, Compression Setting (Sparsity/Bit-width)=W4A162026.0370.51.52
COLAModel=Llama-3.1-8B-Instruct, Calibration Category=Mean2026.034.88—
ZipCalModel=Llama-3.1-8B-Instruct, Calibration Category=Mean2026.034.63-0.25
ZipCalModel=Llama-3.1-8B-Instruct, Compression Method=2SSP, Compression Setting (Sparsity/Bit-width)=25%2026.034.13-0.75
ZipCalModel=gemma-2-9b-it, Compression Method=2SSP, Compression Setting (Sparsity/Bit-width)=25%2026.033.510.21
COLAModel=gemma-2-9b-it, Calibration Category=Mean2026.033.3—
ZipCalModel=gemma-2-9b-it, Calibration Category=Mean2026.032.76-0.54