tinyfl

a tiny federated learning framework built with pytorch and fastapi.

requirements

• python==3.10.10
• torch==1.13.0
• torchvision==0.14.0
• httpx
• uvicorn
• fastapi

installation

use poetry

poetry env use 3.10
poetry install

or

use conda

conda create -n tinyfl python=3.10.10
conda activate tinyfl
pip install torch==1.13.0 torchvision==0.14.0 httpx uvicorn fastapi

quickstart

run the aggregator

poetry run agg configs/agg.config.json

run the parties

poetry run party configs/party0.config.json
poetry run party configs/party1.config.json
poetry run party configs/party2.config.json

get aggregator status

curl {aggregator}/

start training round

curl {aggregator}/start_round

docker

docker build -t tinyfl .
docker run --name agg    --network=host -e CLIENT=0 -e TINYFL_CONFIG=configs/agg.config.json tinyfl:latest
docker run --name party0 --network=host -e CLIENT=1 -e TINYFL_CONFIG=configs/party0.config.json tinyfl:latest
docker run --name party1 --network=host -e CLIENT=1 -e TINYFL_CONFIG=configs/party1.config.json tinyfl:latest
docker run --name party2 --network=host -e CLIENT=1 -e TINYFL_CONFIG=configs/party2.config.json tinyfl:latest