• Docs >
  • Open Source Code Architecture
Shortcuts

Open Source Code Architecture

In general, FedML source code architecture follows the paper which won the Best Paper Award at NeurIPS 2020 (FL workshop). Its original idea is presented at the live video and white paper by FedML co-founder Dr. Chaoyang He.

FedML Code Architecture

Code Architecture

As of March 2022, FedML has been made into an AI company which aims to provide machine learning capability anywhere at any scale. The python version of FedML https://github.com/FedML-AI/FedML-refactor/tree/master/python is now reorganized as follows:

core: The FedML low-level API package. This package implements distributed computing by communication backend like MPI, NCCL, MQTT, gRPC, PyTorch RPC, and also supports topology management. Other low-level APIs related to security and privacy are also supported. All algorithms and Scenarios are built based on the “core” package.

data: FedML will provide some default datasets for users to get started. Customization templates are also provided.

model: FedML model zoo.

device: FedML computing resource management.

simulation: FedML parrot can support: (1) simulate FL using a single process (2) MPI-based FL Simulator (3) NCCL-based FL Simulator (fastest)

cross-silo: Cross-silo Federated Learning for cross-organization/account training

cross-device: Cross-device Federated Learning for Smartphones and IoTs

distributed: Distributed Training: Accelerate Model Training with Lightweight Cheetah

serve: Model serving, tailored for edge inference

mlops: APIs related to machine learning operation platform (open.fedml.ai)

centralized: Some centralized trainer code examples for benchmarking purposes.

utils: Common utilities shared by other modules.

Reference

@article{chaoyanghe2020fedml,
  Author = {He, Chaoyang and Li, Songze and So, Jinhyun and Zhang, Mi and Wang, Hongyi and Wang, Xiaoyang and Vepakomma, Praneeth and Singh, Abhishek and Qiu, Hang and Shen, Li and Zhao, Peilin and Kang, Yan and Liu, Yang and Raskar, Ramesh and Yang, Qiang and Annavaram, Murali and Avestimehr, Salman},
  Journal = {Advances in Neural Information Processing Systems, Best Paper Award at Federate Learning Workshop},
  Title = {FedML: A Research Library and Benchmark for Federated Machine Learning},
  Year = {2020}
}

Video

http://www.youtube.com/watch?v=93SETZGZMyI