GroMo

Gromo is a collaborative effort for designing efficient and growable networks for machine learning. It is a Python package that provides a set of tools for pytorch users to train neural networks, that grow in size and complexity as they learn. The package is designed to be modular and flexible, allowing users to easily design and train their own networks.

The package is built on top of torch.nn.Module, examples shows how to use the package to train a simple neural network, and how to grow the network as it learns. The theoretical and algorithmic details of Gromo are described in the paper indicated below.

The package is still in development, if you would like to contribute, please contact us!

Verbockhaven, M., Rudkiewicz, T., Chevallier, S., and Charpiat, G. (2024). Growing tiny networks: Spotting expressivity bottlenecks and fixing them optimally. Transactions on Machine Learning Research.

Douka, S., Verbockhaven, M., Rudkiewicz, T., Rivaud, S., Landes, F., Chevallier, S., Charpiat, G. (2025). Growth strategies for arbitrary DAG neural architectures. European Symposium on Artificial Neural Networks.

Installation

For users

Install the latest release from PyPI:

pip install gromo

or, in a uv project, uv add gromo.

To install the development version from GitHub:

pip install git+https://github.com/growingnet/gromo.git

For developers

To set up a development environment:

1. Ensure that you have an SSH key set up with GitHub.
2. Fork the repository.
3. Use the following commands, replacing $USER with your GitHub username:

pip install build
git clone git@github.com:$USER/gromo.git --origin origin
cd gromo
git remote add upstream git@github.com:growingnet/gromo.git
git remote set-url --push upstream no-pushing
git fetch upstream
git branch --set-upstream-to=upstream/main main
python -m build
pip install -e .[dev,test]
pre-commit install

Documentation

The documentation for Gromo can be found at: https://growingnet.github.io/gromo/