-
We have provided clear documentation for setting up and using
FedDebug
for anyone. Not only is the codefunctional
, but we have also made extra efforts to make the results ofFedDebug
easily reproducible online. -
Although the local setup is completely functional (details are provided in
INSTALL.md
), we generalizeFedDebug
and make it available for anyone running it onGoogle-Colab
withJupyter Notebooks
to assist users who do not have access to aGPU
for the facilitation of reuse and re-purposing. -
We provide a
FedDebug fault localization technique
in a separate package so that users can use it with any other Federated Learning library with its benchmarks. Furthermore, we provide Docker support to quickly interact withFedDebug Debugging constructs
in no time. -
Complete documentation, including
readme.md
and code comments, is available for reference. -
The
artifact.ipynb
notebook provides a single interface for testing and reproducing the results discussed in theFedDebug
paper for verification and validation purposes.
All of our artifacts can be found at https://github.com/SEED-VT/FedDebug.
We have made every effort to ensure that
FedDebug
is bothfunctional
andreusable
for all users. As the first open-source debugger for federated learning applications with fault localization capabilities, we believe thatFedDebug
will open up new possibilities for debugging and testing in FL in the future.