List your company
ResearchFormerly WizWand

Image Classification on MNIST (Quantization/Bit-width Performance)

98.66Accuracy (w32a32)GPTQ
95.571296.373197.17597.9769Nov 24, 2025
Updated 4mo ago

Evaluation results

MethodLinks
GPTQArchitecture=KAGN Simple2025.1198.6698.6597.9690.1127.3997.9690.1127.39
AdaRoundArchitecture=KAGN Simple2025.1198.6698.7696.5988.4129.8396.5988.4129.83
AWQArchitecture=KAGN Simple2025.1198.6698.797.3557.6513.3697.3557.6513.36
BRECQArchitecture=KAGN Simple2025.1198.6698.7396.1176.0723.3396.2281.712.8
GPTQArchitecture=KAN FCN2025.1197.999897.8697.0776.5895.8888.9940.9
AdaRoundArchitecture=KAN FCN2025.1197.9997.9797.7896.3642.8795.7588.8832.35
AWQArchitecture=KAN FCN2025.1197.9997.9897.895.5333.6595.5588.7321.02
BRECQArchitecture=KAN FCN2025.1197.9973.7772.5873.650.5170.8786.9352.38
GPTQArchitecture=KAN ConvNet2025.1195.6995.6995.6595.6388.1810.110.110.1
AdaRoundArchitecture=KAN ConvNet2025.1195.6995.6995.6195.280.039.7310.2810.25
AWQArchitecture=KAN ConvNet2025.1195.6995.6895.6295.0162.710.110.110.32
BRECQArchitecture=KAN ConvNet2025.1195.6936.3737.7930.3135.0775.693.5264.97