prunhild is a small library for neural network pruning based on PyTorch.
I wrote this library for better structuring my code when replicating the experiments from The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks for a university course.
The library currently only implements simple magnitude-based pruning as used in the previously mentioned paper. Througout the library magnitude-pruning is referred to as cutoff-pruning.
The library offers a simple class CutoffPruner which offers a similar interface as the Optimizer class from PyTorch. The Pruner class takes different strategies (Cutoffs) for computing the binary masks (prune-masks) used for pruning the weights.
You can install the library directly from GitHub:
# Install from GitHub
pip install git+https://github.com/gfrogat/prunhild
The folder examples contains an example using MNIST on how to use the library. If you encounter problems or have any questions don't hesitate to open an issue.
- prunhild example on MNIST - code