Installing FedML

FedML supports Linux, MacOS, Windows, and Android.

FedML Source Code Repository

https://github.com/FedML-AI/FedML

Installing with pip

pip install fedml

if your machine has not installed python, please install one version of the following pythons: 3.8, 3.9, 3.10.

Installing with pip on Ubuntu

On Ubuntu, run the following commands to install pip3 and fedml.

sudo apt install python3-pip
pip3 install fedml
export PATH=$HOME/.local/bin:$PATH
fedml env

Installing with pip on CentOS

On CentOS, run the following commands to install pip3 and fedml.

yum –y install python3-pip
pip3 install fedml
export PATH=$HOME/.local/bin:$PATH
fedml env

The default machine learning engine is PyTorch. FedML also supports TensorFlow, Jax, and MXNet. You can install related engines as follows:

pip install "fedml[MPI]"
pip install "fedml[tensorflow]"
pip install "fedml[jax]"
pip install "fedml[mxnet]"

For MPI installation, it’s used for local distributed simulation with MPI (https://mpi4py.readthedocs.io/en/stable/). On MacOS, the installation commands in conda environment is:

conda install mpi4py openmpi

About OpenMPI library installation for MPI, the reference is as follows: https://docs.open-mpi.org/en/v5.0.x/installing-open-mpi/quickstart.html For OpenMPI on MacOS, please review the following links: https://betterprogramming.pub/integrating-open-mpi-with-clion-on-apple-m1-76b7815c27f2 https://formulae.brew.sh/formula/open-mpi

The above commands work properly in Linux environment. For Windows/Mac OS (Intel)/Mac OS (M1), you may need to follow TensorFlow/Jax/MXNet official guidance to fix related installation issues.

Installing FedML with Anaconda

conda create --name fedml-pip python=3.8
conda activate fedml-pip
conda install --name fedml-pip pip
pip install fedml

(Note: please use python 3.8 if you met any compatability issues. Currently, we support 3.7, 3.8, 3.9, 3.10, 3.11. on MacOS and Python 3.11, if you meet any issues, please run the command: brew install autoconf automake libffi libtool pkg-config, then run the command to install fedml again: pip install fedml)

After installation, please use “pip list | grep fedml” to check whether fedml is installed.

Compatibility with HomeBrew-Installed Python On Apple Silicon Mac

If you are using Apple Silicon MAC, we suggest using Conda to install python 3.8+ and related lib on your device.

But if you have used HomeBrew to install python, and having problem with running “pip install fedml” command, in this case you need to ensure:

a. Two environment path on your device need to be specify to use Conda not HomeBrew:

(1) In ~/.bash_profile and ~/.zprofile, the path to python bin file, need to be Conda python file location.

(2) In ~/.bash_profile and ~/.zprofile, the path to pip bin file, need to be Conda pip file location.

b. When you encounter with C/C++ compiler issue, try:

(3) conda install …

Installing FedML from Debugging and Editable Mode

cd python
pip install -e ./

Installing FedML from Source

git clone https://github.com/FedML-AI/FedML.git && \
cd ./FedML/python && \
python setup.py install

If you want to run examples with TensorFlow, Jax or MxNet, you need to install optional dependencies:

git clone https://github.com/FedML-AI/FedML.git && \
cd ./FedML/python && \
python install -e '.[tensorflow]'
python install -e '.[jax]'
python install -e '.[mxnet]'

(Notes: Tensorflow example located in tf_mqtt_s3_fedavg_mnist_lr_example directory, Jax example location in jax_haiku_mqtt_s3_fedavg_mnist_lr_example directory)

If you need to install from a specific commit (normally used for the debugging/development phase), please follow commands below:

git clone https://github.com/FedML-AI/FedML.git && \
cd ./FedML && git checkout e798061d62560b03e049d514e7cc8f1a753fde6b && \
cd python && \
python setup.py install

Please change the above commit id to your own (you can find it at https://github.com/FedML-AI/FedML/commits/master)

Running FedML in Docker (Recommended)

FedML Docker Hub: https://hub.docker.com/repository/docker/fedml/fedml

We recommend using FedML in the Docker environment as it circumvents complex and tedious installation debugging. Currently, we maintain docker images for x86_64 architecture. But for your own purpose, you may build your docker image to support the following architectures: arm, raspberrypi, nvidia jetson via our Dockerfile located in the directory https://github.com/FedML-AI/FedML/tree/master/installation/build_fedml_docker

Please refer to the following commands and remember to change LOCAL_WORKSPACE to your own.

FedML Standard Docker Image

The FedML standard docker image can support to run on CPU an GPU devices. It deviated from the Nvidia official image which is large size. So the FedML standard docker image will be a large image. Now it is about 17GB in size. Up to now, the FedML standard docker image can run on the Linux platform. If you want to run on the MacOS platform, you should use the FedML light docker image which can be running on multiple architectures, e.g. X86, ARM, etc.

(1) Pull the standard Docker image and prepare the docker environment

FEDML_DOCKER_IMAGE=fedml/fedml:latest-torch1.13.1-cuda11.6-cudnn8-devel
docker pull $FEDML_DOCKER_IMAGE

# if you want to use GPUs in your host OS, please follow this link: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker
sudo apt-get update
sudo apt-get install -y nvidia-docker2
sudo systemctl restart docker
sudo chmod 777 /var/run/docker.sock

(2) Run standard Docker with interactive mode

On GPUs:

FEDML_DOCKER_IMAGE=fedml/fedml:latest-torch1.13.1-cuda11.6-cudnn8-devel
LOCAL_WORKSPACE=$PleaseUseYourLocalDirectory
DOCKER_WORKSPACE=/home/fedml/fedml_source

docker run -v $LOCAL_WORKSPACE:$DOCKER_WORKSPACE --shm-size=64g --ulimit nofile=65535 --ulimit memlock=-1 --privileged --gpus all --network=host --env WORKSPACE=$DOCKER_WORKSPACE -ti $FEDML_DOCKER_IMAGE /bin/bash

On CPUs:

FEDML_DOCKER_IMAGE=fedml/fedml:latest-torch1.13.1-cuda11.6-cudnn8-devel
LOCAL_WORKSPACE=$PleaseUseYourLocalDirectory
DOCKER_WORKSPACE=/home/fedml/fedml_source

ddocker run -v $LOCAL_WORKSPACE:$DOCKER_WORKSPACE --shm-size=64g --ulimit nofile=65535 --ulimit memlock=-1 --privileged --network=host --env WORKSPACE=$DOCKER_WORKSPACE -ti $FEDML_DOCKER_IMAGE /bin/bash

You should now see a prompt that looks something like, you may run the ‘fedml login $YourUserId’ to log into the MLOps platform.

root@142ffce4cdf8:/#
root@142ffce4cdf8:/# fedml login 1606

And also, you may enter into the $WORKSPACE which is your host directory to run your own examples:

root@142ffce4cdf8:/#
root@142ffce4cdf8:/# cd $WORKSPACE
root@142ffce4cdf8:/home/fedml/fedml_source#

FedML Light Docker Image

(1) Run light Docker with interactive mode

The light docker is a smaller image about 2.3GB size. So it can pull and run more smoothly. The light docker just supports cpu arch. So, if you want to use the GPU, you should use the above standard Docker with gpu options. Each docker image needs more than 5GB memory size to run the fedml learning task. (This is estimated with the MNist dataset, if you use other dataset, Maybe the memory size is larger or smaller than the size with the MNist dataset) So, you need to reserve sufficient memory size for your federated learning task. On MacOS, you should set memory size in the navigation path DockerDesktop -> Preference -> Resource -> Memory. If you want to run three docker containers simultaneously, you need to set the resource memory to not less than 15GB.

FEDML_DOCKER_IMAGE=fedml/fedml:light
LOCAL_WORKSPACE=$PleaseUseYourLocalDirectory
DOCKER_WORKSPACE=/home/fedml/fedml_source

docker run -v $LOCAL_WORKSPACE:$DOCKER_WORKSPACE --shm-size=64g --ulimit nofile=65535 --ulimit memlock=-1 --privileged --network=host --env WORKSPACE=$DOCKER_WORKSPACE -ti $FEDML_DOCKER_IMAGE /bin/bash

You should now see a prompt that looks something like, you may run the ‘fedml login $YourUserId’ to log into the MLOps platform.

root@142ffce4cdf8:/#
root@142ffce4cdf8:/# fedml login 1606

And also, you may enter into the $WORKSPACE which is your host directory to run your own examples:

root@142ffce4cdf8:/#
root@142ffce4cdf8:/# cd $WORKSPACE
root@142ffce4cdf8:/home/fedml/fedml_source#

(2) Run light Docker with daemon mode and automatically log into the MLOps platform

You may run the light docker as the daemon mode and automatically log into the MLOps platform as the client. The commands ars as follows:

FEDML_DOCKER_IMAGE=fedml/fedml:light
LOCAL_WORKSPACE=$PleaseUseYourLocalDirectory
DOCKER_WORKSPACE=/home/fedml/fedml_source
YOUR_FEDML_USER_ID=1606

docker run -v $LOCAL_WORKSPACE:$DOCKER_WORKSPACE --shm-size=64g --ulimit nofile=65535 --ulimit memlock=-1 --privileged --network=host --env WORKSPACE=$DOCKER_WORKSPACE -d $FEDML_DOCKER_IMAGE bash -c 'fedml login '$YOUR_FEDML_USER_ID';sleep 100000'

You may run the light docker as the daemon mode and automatically log into the MLOps platform as the server with the option ‘-s’. The commands ars as follows:

FEDML_DOCKER_IMAGE=fedml/fedml:light
LOCAL_WORKSPACE=$PleaseUseYourLocalDirectory
DOCKER_WORKSPACE=/home/fedml/fedml_source
YOUR_FEDML_USER_ID=1606

docker run -v $LOCAL_WORKSPACE:$DOCKER_WORKSPACE --shm-size=64g --ulimit nofile=65535 --ulimit memlock=-1 --privileged --network=host --env WORKSPACE=$DOCKER_WORKSPACE -d $FEDML_DOCKER_IMAGE bash -c 'fedml login -s '$YOUR_FEDML_USER_ID';sleep 100000'

After you run the above command, the terminal will show the container id like the following format. b0769135f8e65c5b0b7b7cb9666f3f910a4e431c25084ed72ae059ea1a6376af

If you want to show logs for the fedml light container, you may run the following command with the above container id.

docker logs b0769135f8e65c5b0b7b7cb9666f3f910a4e431c25084ed72ae059ea1a6376af

If you want to list the fedml light containers, you may run the command.

docker ps |grep fedml:light

If you want to kill all fedml light containers, the command is as follows.

docker stop `docker ps |grep fedml:light |awk -F' ' '{print $1}'`

(4) Run the interpreter in PyCharm or Visual Studio using Docker environment

• PyCharm

https://www.jetbrains.com/help/pycharm/using-docker-as-a-remote-interpreter.html#summary

• Visual Studio

https://code.visualstudio.com/docs/remote/containers

(4) Other useful commands

# docker rm $(docker ps -aq)
docker container kill $(docker ps -q)

Running FedML on Kubernetes

This tutorial will guide you to deploy your fedml client and server to target Kubernetes pods running on GPU/CPU physical nodes.

The entire workflow is as follows:

(k8s deployment file is located in: https://github.com/FedML-AI/FedML/tree/master/installation/install_on_k8s/fedml-edge-client-server)

1. In the file fedml-edge-client-server/deployment-client.yml, modify the variable ACCOUNT_ID to your desired value

2. Deploy the fedml client: kubectl apply -f ./fedml-edge-client-server/deployment-client.yml

3. In the file fedml-edge-client-server/deployment-server.yml, modify the variable ACCOUNT_ID to your desired value

4. Deploy the fedml server: kubectl apply -f ./fedml-edge-client-server/deployment-server.yml

5. Login the FedML MLOps platform (https://open.fedml.ai), the above deployed client and server will be found in the edge devices

If you want to scale up or scal down the pods to your desired count, you may run the following command:

kubectl scale -n $YourNameSpace --replicas=$YourDesiredPodsCount deployment/fedml-client-deployment

kubectl scale -n $YourNameSpace --replicas=$YourDesiredPodsCount deployment/fedml-server-deployment

Installation with Helm Charts

Also, you may use the helm charts to deploy your fedml client and server to target Kubernetes cluster. You just need to run the following commands with your user id at the open.fedml.ai.

kubectl create namespace fedml
helm install --set image.repository="fedml/fedml-edge-client-server-light" --set env.fedmlAccountId="$YourUserId" --set env.role="client" fedml-client-deployment ./fedml-client-deployment-latest.tgz
helm install --set image.repository="fedml/fedml-edge-client-server-light" --set env.fedmlAccountId="$YourUserId" --set env.role="server" fedml-server-deployment ./fedml-server-deployment-latest.tgz

Q&A

1. Q: How to scale up or scale down?
   A: Use the following commands:

kubectl scale -n $YourNameSpace --replicas=$YourDesiredPodsCount deployment/fedml-client-deployment

kubectl scale -n $YourNameSpace --replicas=$YourDesiredPodsCount deployment/fedml-server-deployment

2. Q: FedML Client send online status to FedML Server via which protocol?
   A: Via MQTT

3. Q: FedML Client send model, gradient parameters to FedML Server via which protocol?
   A: Use S3 protocol to store and exchange models and use MQTT to exchange messages between FedML Client and Server

4. Q: Why do we need AWS S3?
   A: Use S3 protocol to store and exchange models.

Guidance for Windows Users

Please follow instructions at Windows Installation

Guidance for Raspberry Pi Users

Please follow instructions at Raspberry Pi Installation

Guidance for NVIDIA Jetson Devices

Please follow instructions at NVIDIA Jetson Device Installation

Testing if the installation succeeded

If the installation is successful, you will not see any issue when run import fedml.

(mnn37) chaoyanghe@Chaoyangs-MBP FedML-refactor % python
Python 3.7.7 (default, Mar 26 2020, 10:32:53) 
[Clang 4.0.1 (tags/RELEASE_401/final)] :: Anaconda, Inc. on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import fedml
>>> 

Installing FedML Android SDK/APP

Please follow the instructions at https://github.com/FedML-AI/FedML/java/README.md

Troubleshooting

If you met any issues during installation or have additional installation requirements, please post issues at https://github.com/FedML-AI/FedML/issues