finds.dev

FedML-AI/FedML

FEDML - The unified and scalable ML library for large-scale distributed training, model serving, and federated learning. FEDML Launch, a cross-cloud scheduler, further enables running any AI jobs on any GPU cloud or on-premise cluster. Built on this library, TensorOpera AI (https://TensorOpera.ai) is your generative AI platform at scale.

View on GitHubHomepage ↗

★ 4,048PythonApache-2.0updated Oct 2025

// the take

FedML is a federated learning and distributed training framework that lets you run ML jobs across edge devices, on-prem clusters, and multiple clouds. It's backed by a commercial platform (TensorOpera AI) and targets researchers and MLOps teams who need to train models on data that can't leave its source. The scope is genuinely broad: cross-device mobile training, cross-silo server federation, simulation, and LLM fine-tuning.

The federated learning core is mature — FedAvg, differential privacy, secure aggregation, and Byzantine-robust aggregation are all implemented and CI-tested via separate smoke test workflows. The Android SDK with on-device training via MNN (Alibaba's inference engine) is a real differentiator; very few open-source projects actually support model training on mobile hardware. The simulation mode using MPI lets you prototype federations locally without spinning up actual distributed infrastructure. CI is thorough: separate workflows per algorithm variant (LightSecAgg, CDP, LDP, attack/defense) rather than one catch-all test.

The project has essentially become a marketing funnel for TensorOpera.ai — the README is mostly about the commercial platform, not the open-source library. Documentation for the OSS path specifically is thin and hard to find amid the product promotion. The repo is enormous and unfocused: Android Java, C++ JNI with MNN, Python training code, YAML job configs, and MLOps orchestration are all mixed together with no clear separation of concerns, making it hard to take just one piece. Activity has slowed since early 2024 and the last push being October 2025 with sparse commit cadence suggests the team's energy is on the SaaS product, not the open-source core.

We read the README, the directory tree and a key source file in June 2026. Nobody pays to be listed. How we judge →

// at a glance

Stars
4,048
Forks
766
Language
Python
License
Apache-2.0
Last commit
28 Oct 2025

// nearby

If FedML isn't quite it.

All repos →