Ray is Python-native. Scale and distribute any Python code for use cases like simulation, backtesting, and more.

Process structured and unstructured data including images, videos, audio, and more.

Run distributed training, including Gen AI foundation models, time series models, and traditional AI / ML models like XGBoost at scale with 1 line of code.  And yes – it’s compatible with your framework of choice.

Deploy models and business logic - not instances. Ray Serve offers independent scaling and fractional resources to let you get the most of the models you deploy. Get support for any ML model – from LLMs to stable diffusion models to object detection models and beyond.

Leverage heterogeneous compute to streamline offline batch inference workflows. Use CPUs and GPUs in the same pipeline to increase utilization, fully saturate GPUs, and decrease costs.

Run best-in-class reinforcement learning workflows. Ray RLlib supports production-level, highly distributed RL workloads while maintaining unified and simple APIs for a large variety of industry applications.

Build end-to-end GenAI workflows with Ray. Ray supports multimodal models, RAG applications, and more.

Serve Large Language Models and scale seamlessly with Ray. Ray's flexility to support any accelerator, any model, coupled with seamless scaling is built for LLM inference (online and batch).

From the framework behind ChatGPT, easily fine tune Large Language models at scale.