About Ray
Ray is an open source framework for running Python code and AI workloads across one machine or a large cluster. It is developed by Anyscale and is used for training, serving and data processing jobs.
What it does
Ray lets a developer take Python code that runs on a laptop and spread it across many CPUs and GPUs. Its libraries cover common AI jobs: distributed model training, serving models behind an endpoint, batch inference over large datasets, and processing of image, video and audio files. Scaling is set in code, so the same program can start small and grow.
Key features
- Run ordinary Python code across many machines
- Distributed training for machine learning models
- Model serving with independent scaling per model
- Batch inference using CPUs and GPUs in one pipeline
- Processing of structured and unstructured data
Who it's for
Machine learning engineers, data teams and backend developers who run compute-heavy Python jobs and want them to scale without rewriting their code. It is also used by teams that serve several different types of model in production.
Pricing and good to know
Ray itself is open source under the Apache-2.0 licence and is free to install. Running it on a cluster means paying for the machines, whether they are your own or a cloud provider's. Anyscale also sells a managed, enterprise version of Ray, which is separate from the open source project.
Highlights
- Run ordinary Python code across many machines
- Distributed training for machine learning models
- Model serving with independent scaling per model
- Batch inference using CPUs and GPUs in one pipeline
- Processing of images, video, audio and other data
