Fed ML

Founded by Salman Avestimehr in 2022

Train and deploy machine learning models across decentralized edge devices and cloud servers

Pricing model

Paid

Rating

★ Mixed Reviews

Time to value

Moderate Setup (1-3 hours)

You can use FedML to build, train, and deploy machine learning models without moving sensitive data to centralized servers. The system coordinates federated learning across edge devices like smartphones, IoT hardware, and multi-cloud environments while keeping raw user data private. You can access on-demand GPU clusters for training, deploy open-source models using serverless API endpoints, and host custom models on dedicated hardware. It also provides pre-built templates for fine-tuning models, building retrieval-augmented generation agents, and managing cross-silo distributed training workflows.

What Fed ML does

Run federated learning algorithms across smartphone fleetsLaunch distributed model training jobs on decentralized GPUsDeploy open-source LLMs through serverless API endpointsConfigure autoscaling dedicated GPU endpoints for custom modelsBuild conversational agents connected to external vector databasesExecute client-side training sessions inside web browsersMonitor hardware usage and latency across distributed computing nodesTrains models directly on iOS and Android devices without extracting raw user dataCoordinates cross-silo distributed training across isolated clouds and edge serversRuns browser-based training using client-side JavaScript executionProvides serverless pay-per-token API endpoints for open-source foundation modelsHosts dedicated model endpoints with autoscaling and version trackingMatches machine learning training jobs with low-cost decentralized GPU resourcesSupports custom retrieval-augmented generation agents with vector databases