Where-to-Prune

Using LSTM to guide filter-level pruning. LSTM is employed as an evaluation metric to generate the pruning decision for each conv layer.

Installation

To run this script, you need Pytorch and CUDA. This code is written in Pytorch 3.5.

Running

To run the script with default parameters,

python train.py

model0.pkl is a baseline model of VGG19 on CIFAR-10 dataset. This script will prune model0.pkl and generate a slimmer model.