Unofficial implementation of SNIP (ICLR 19) in PyTorch. SNIP is a single shot neural network prunning technique which prunes the network before training based on sensitivity of connections of the randomly initialized weights.

from snip_prunner import Prunner
from model import my_model
from loss_func import my_loss

prunner = Prunner(my_model, my_loss, train_dataloader)
prunned_model, masks = prunner.prun(compression_factor=0.9, num_batch_sampling=1)

"""
Now continue training prunned_model 
as you would do in normal setup
"""

Refer test_mnist.ipynb for experiments on MNIST

Run experiments using ResNet Model on CIFAR 10

SNIP