stijnverdenius / compression_tutorial

Pruning tutorial

This repository contains a small tutorial on pruning. We are gonna train a simple feed forward network classifier on a toy dataset with two classes and features of dim=128. Each epoch we will prune, using a either structured layer-wise magnitude or unstructured globally magnitude criterion

• main.py defines the training and pruning schedule
• model.py defines the Network
• PrunableModel.py contains a class from which a model can inherit, such that it will create the needed variables for pruning. Additionally it has all the required functions.
• ./logs will be a folder where tensorboard files will be generated. These function as logs and can be viewed by running tensorboard --logdir ./logs