Pybiscus is a simple tool to perform Federated Learning on various models and datasets. It aims at automated as much as possible the FL pipeline, and allows to add virtually any kind of dataset and model.
Pybiscus is built on top of Flower, a mature Federated Learning framework; Typer (script and CLI parts) and Lightning/Fabric for all the Machine Learning machinery.
It is managed using the uv package manager
extend your user's path to include the bin directory download the required packages
uv sync
You can find example launch scripts in ./launch/uv subdirectories
extend your user's path to include the bin directory produce the image
cd ./container
./build_pybiscus_container.sh
You can find example launch scripts in ./launch/container subdirectories
Documentation is available at docs.
If you are interested in contributing to the Pybiscus project, start by reading the Contributing guide.
Pybiscus is on active development at Thales, both for internal use and on some collaborative projects. One major use is in the Europeean Project PAROMA-MED, dedicated to Federated Learning in the context of medical data distributed among several Hospitals.
The License is Apache 2.0. You can find all the relevant information here LICENSE