Federated learning workshop

Day 2 practical

python client.py --partition-id 0

Please replace 0 with a number 0-5

Software installation

It would be helpful if you could install the following software before the workshop. If you have any trouble, we can help you during the workshop.

Installing Docker

Can be done with instructions from https://www.docker.com/get-started/ And docker compose https://docs.docker.com/compose/install/linux/

Using a Docker image

This is the easiest way to get started. You can run the following command to pull the Docker image and run a container. The main benefit of this method is that it hides all the painful (Nvidia and PyTorch) installation details from you. Since this is relatively large, please let me know if you need help downloading it. The container image is self-hosted and available here. To download this image you might need the Tor Browser.

Building your local container

If you prefer to build your container, use the following Dockerfile. This will install all the required software in a container.

# Description: Dockerfile for the hereditary project
#Start from the base pytorch image
ARG PYTORCH_IMAGE=nvcr.io/nvidia/pytorch:23.12-py3
FROM ${PYTORCH_IMAGE}

#Set the working directory
WORKDIR /mnt/workspace/

RUN python3 -m pip install -U pip
RUN python3 -m pip install -U setuptools

# RUN python3 -m pip install torch torchvision torchaudio
#Install the required packages
#General python packages

RUN python3 -m pip install matplotlib transformers evaluate datasets scikit-learn tqdm pillow pytorch_lightning jupyter notebook
#FL frameworks
##NVFlare
# RUN git clone https://github.com/NVIDIA/NVFlare.git --branch ${NVF_BRANCH} --single-branch NVFlare
# RUN cd NVFlare/
# RUN python3 setup.py install
# RUN cd -
RUN python3 -m pip install nvflare
##Flower
RUN python3 -m pip install flwr flwr_datasets
##Pysyft
# RUN git clone https://github.com/OpenMined/PySyft.git
# RUN cd PySyft
# RUN python3 setup.py install
# RUN python3 setup.py test
# RUN cd -
RUN python3 -m pip install syft

You can build the container using the following command

docker build -t fedai .

You can run the container using the following command

docker run -it --rm --gpus all -v $(pwd):/mnt/workspace/ fedai

Using a virtual environment

If you prefer to install the software on your local machine, follow the instructions below. This method is more flexible and allows you to customize the software installation. It also relies on you setting up all your (GPU) drivers and libraries correctly.

First, create a virtual environment.

python3 -m venv fedai

Then, activate the virtual environment.

source fedai/bin/activate

Then install the required software

python3 -m pip install -U pip
python3 -m pip install -U setuptools
python3 -m pip install matplotlib scikit-learn torch torchvision torchaudio transformers evaluate datasets tqdm pillow pytorch_lightning jupyter notebook
python3 -m pip install nvflare
python3 -m pip install flwr flwr-datasets
python3 -m pip install syft