CodeFlare provides a simple, user-friendly abstraction for developing, scaling, and managing resources for distributed AI/ML on the Hybrid Cloud platform with OpenShift Container Platform.
CodeFlare stack consists of the following main components. This project is organized as a metarepo, gathering pointers and artifacts to deploy and use the stack.
-
Simplified user experience: CodeFlare SDK and CLI to define, develop, and control remote distributed compute jobs and infrastructure from either a python-based environment or command-line interface
-
Efficient resource management: Multi-Cluster Application Dispatcher (MCAD) for queueing, resource quotas, and management of batch jobs. And Instascale for on-demand resource scaling of an OpenShift cluster
-
Automated and streamlined deployment: CodeFlare Operator for automating deployment and configuration of the Project CodeFlare stack
With CodeFlare stack, users automate and simplify the execution and scaling of the steps in the life cycle of model development, from data pre-processing, distributed model training, model adaptation and validation.
Through transparent integration with Ray and PyTorch frameworks, and the rich library ecosystem that run on them, CodeFlare enables data scientists to spend more time on model development and minimum time on resource deployment and scaling.
See below our stack and how to get started.
In addition to running standalone, Project CodeFlare is deployed as part of and integrated with the Open Data Hub, leveraging OpenShift Container Platform.
With OpenShift, CodeFlare can be deployed anywhere, from on-prem to cloud, and integrate easily with other cloud-native ecosystems.
Watch this video for an introduction to Project CodeFlare and what the stack can do.
To get started using the Project CodeFlare stack, try this end-to-end example!
For more basic walk-throughs and in-depth tutorials, see our demo notebooks!
See more details in any of the component repos linked above, or get started by taking a look at the project board for open tasks/issues!
We attempt to document all architectural decisions in our ADR documents. Start here to understand the architectural details of Project CodeFlare.
Join our Slack community to get involved or ask questions.
CodeFlare related blogs are published on our Medium publication.
CodeFlare is an open-source project with an Apache 2.0 license.