Post-training static quantization using ResNet18 architecture
pytorch
mnist-classification
quantization
resnet-18
post-training-quantization
floatfunctional
fbgemm
float32-to-int8
reduce-inference-time
reduce-memory-storage
-
Updated
Aug 1, 2020 - Jupyter Notebook