This is the official repository to the ICLR 2021 paper "ChipNet: Budget-Aware Pruning with Heaviside Continuous Approximations" by Rishabh Tiwari, Udbhav Bamba, Arnav Chavan, Deepak Gupta.
You will need Python 3.7 and the packages specified in requirements.txt. We recommend setting up a virtual environment with pip and installing the packages there.
Install packages with:
$ pip install -r requirements.txt
All configurations concerning data, model, training, etc. can be called using commandline arguments.
The main script offers many options; here are the most important ones:
usage: python pretraining.py [dataset] [model name] --epochs [number of epochs] --decay [weight decay] --batch_size [batch size] --lr [learning rate] --scheduler_type {1, 2}
usage: python pruning.py [dataset] [model name] --Vc {float value between 0-1} --budget_type {channel_ratio, volume_ratio, parameter_ratio, flops_ratio} --epochs 20
usage: python finetuning.py [dataset] [model name] --Vc {float value between 0-1} --budget_type {channel_ratio, volume_ratio, parameter_ratio, flops_ratio} --name {model name}_{dataset}_{budget}_{budget_type} --epochs [number of epochs] --decay [weight decay] --batch_size [batch size] --lr [learning rate] --scheduler_type {1, 2}
usage: python finetuning.py [dataset] [model name] --Vc {float value between 0-1} --budget_type {channel_ratio, volume_ratio, parameter_ratio, flops_ratio} --host_name {model name}_{dataset}_{budget}_{budget_type} --epochs [number of epochs] --decay [weight decay] --batch_size [batch size] --lr [learning rate] --scheduler_type {1, 2}
- Parameter and FLOPs budget is supported only with models using ResNetCifar module for now.
Please cite our paper in your publications if it helps your research. Even if it does not,and you want to make us happy, do cite it :)
@inproceedings{
tiwari2021chipnet,
title={ChipNet: Budget-Aware Pruning with Heaviside Continuous Approximations},
author={Rishabh Tiwari and Udbhav Bamba and Arnav Chavan and Deepak Gupta},
booktitle={International Conference on Learning Representations},
year={2021},
url={https://openreview.net/forum?id=xCxXwTzx4L1}
}
This project is licensed under the MIT License.