-- update on Nov 26th
Launch streamlit:
streamlit run app.pyTODO:
- add rerun & stop
- test every epoch and show on figure
- auto select sparse ratio and method
- more dashboard windows
- [ ]
- I modified TorchVision official implementation of popular CNN models, and trained those on CIFAR-10 dataset. Also used Neural Network Intelligence(https://nni.readthedocs.io/en/stable/compression/overview.html) for pruning.
- The weights of these models are also shared so you can just load the pre trained models.
- Start by installing the requirements file
pip install -r requirements.txt
Automatically download and extract the weights from Box (933 MB)
python train.py --download_weights 1Check the train.py to see all available hyper-parameter choices.
python train.py --classifier resnet18