Official PyTorch implementation of LilNetX: Lightweight Networks with EXtreme Model Compression and Structured Sparsification

Create new virtualenv/conda environment and run the following command

pip install -r requirements.txt

To run ImageNet experiments with the faster FFCV implementation, follow instructions on https://github.com/libffcv/ffcv to install the FFCV library.

Hyperparameters are defined in yaml files in the configs folder. An example run command for training a ResNet20 with width 4 on the Cifar10 dataset would look like

python main.py --config configs/cifar10_resnet20.yaml

Hyperparameters can additionally be overriden with command-line arguments which are dot separated. For e.g. training the same model as above but with train batch size of 512 (instead of the default 256) would look like

python main.py --config configs/cifar10_resnet20.yaml --trainer.train_batch 512

This project is released under the MIT License. Please review the License file for more details.