Follow these repo instructions to set up your environment, paths, datasets, number of works, etc.:
https://github.com/facebookresearch/open_lth
https://github.com/sahibsin/Pruning
Run dnnshifter-training.sh
to train a portfolio of sparse models, then run othermethods-training-cifar10.sh
/othermethods-training-tinyimagenet.sh
to train the other methods (Random, Magnitude, Synflow etc.)
or
Download pretained sparse models here
Follow this notebook on how to use DNNShifter to prune sparse models structurally.
(Note: Update paths where prompted, and models need to be stored in this directory structure format <method_name>/<pruning_level>/<.pth file>
)