Experiments for the paper "4.6-bit Quantization for Fast and Accurate Neural Network Inference on CPUs"

by Anton Trusov, Elena Limonova, Dmitry Nikolaev, and Vladimir V. Arlazarov

This reposetry contain the sorce code for trainng 4.6-bit quantized network in pytorch, as it was used in the experiments for the paper "4.6-bit Quantization for Fast and Accurate Neural Network Inference on CPUs".

To directly reproduce the experiments run one of the following commands

• python3 experiment1_cifar.py -m CNN[6-10] -p [prexix] --step_epochs 50 --lr_scale 0.5 --n_exp_rep 5

• python3 experiment2_resnet18.py -p [prefix] --n_exp_rep 5

• python3 experiment2_resnet34.py -p [prefix] --n_exp_rep 5

This will save results in .json format in the ./results subdirectory.

[prefix] may be any string. it is used as a prefix in result file names.

For ResNet experiments you might want to provide paths to training and validation sets, using --imagenet_train_path and --imagenet_val_path options.