Ray.io
An AI compute engine that consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.
Ray is a unified framework for scaling AI and Python applications. Ray consists of a core distributed runtime and a set of AI libraries for simplifying ML compute:
Learn more about Ray AI Libraries:
- Data: Scalable Datasets for ML
- Train: Distributed Training
- Tune: Scalable Hyperparameter Tuning
- RLlib: Scalable Reinforcement Learning
- Serve: Scalable and Programmable Serving
Or more about Ray Core and its key abstractions:
- Tasks: Stateless functions executed in the cluster.
- Actors: Stateful worker processes created in the cluster.
- Objects: Immutable values accessible across the cluster.
Learn more about Monitoring and Debugging:
- Monitor Ray apps and clusters with the Ray Dashboard.
- Debug Ray apps with the Ray Distributed Debugger.
Ray runs on any machine, cluster, cloud provider, and Kubernetes, and features a growing ecosystem of community integrations.
Install Ray with: pip install ray. For nightly wheels, see the Installation page.
Why Ray?
Today's ML workloads are increasingly compute-intensive. As convenient as they are, single-node development environments such as your laptop cannot scale to meet these demands.
Ray is a unified way to scale Python and AI applications from a laptop to a cluster.
With Ray, you can seamlessly scale the same code from a laptop to a cluster. Ray is designed to be general-purpose, meaning that it can performantly run any kind of workload. If your application is written in Python, you can scale it with Ray, no other infrastructure required.
More Information
- Documentation
- Ray Architecture whitepaper
- Exoshuffle: large-scale data shuffle in Ray
- Ownership: a distributed futures system for fine-grained tasks
- RLlib paper
- Tune paper
Older documents:
Libraries
New Libraries
This section contains libraries that are well-made and useful, but have not necessarily been battle-tested by a large userbase yet.
Models and Projects
Ray + LLM
- veRL veRL: Volcano Engine Reinforcement Learning for LLM
- FastChat Vicuna: An Open-Source Chatbot Impressing GPT-4 with 90% ChatGPT Quality
- LangChain-Ray Examples on how to use LangChain and Ray
- Aviary Ray Aviary - evaluate multiple LLMs easily
- LLM-distributed-finetune Finetuning Large Language Models Efficiently on a Distributed Cluster, Uses Ray AIR to orchestrate the training on multiple AWS GPU instances.
- LLMPerf - A library for validating and benchmarking LLMs (updated through 2024)
Reinforcement Learning
- slime - A LLM post-training framework aiming at scaling RL.
- muzero-general - A commented and documented implementation of MuZero based on the Google DeepMind paper (Schrittwieser et al., Nov 2019) and the associated pseudocode.
- rllib-torch-maddpg - PyTorch implementation of MADDPG (Lowe et al.) in RLLib
- MARLlib - a comprehensive Multi-Agent Reinforcement Learning algorithm library
- VMAS - A vectorized differentiable simulator for Multi-Agent Reinforcement Learning benchmarking
Ray Data (Data Processing)
- RayDP - Distributed data processing library on Ray by running Apache Spark on Ray. Seamlessly integrates with other Ray libraries for E2E data analytics and AI pipeline.
- Google Cloud Platform Ray Preprocessing - Examples of Ray data preprocessing pipelines for model fine-tuning on GCP.
Ray Train (Distributed Training)
- Ray Train Examples - Official Ray Train documentation with PyTorch, TensorFlow, and Hugging Face Accelerate examples for distributed training.
- MinIO with Ray Train - Distributed training examples using Ray Train with MinIO object storage.
Ray Tune (Hyperparameter Optimization)
- Ultralytics YOLO11 with Ray Tune - Efficient hyperparameter tuning for YOLO11 object detection models using Ray Tune.
- Softlearning - Reinforcement learning framework for training maximum entropy policies, official implementation of Soft Actor-Critic algorithm using Ray Tune.
- Flambe - ML framework to accelerate research and its path to production, integrates with Ray Tune.
Ray Serve (Model Serving)
- LangChain Ray Serve - Deploy LangChain applications and OpenAI chains in production using Ray Serve.
Ray + JAX / TPU
- Swarm-jax - Swarm training framework using Haiku + JAX + Ray for layer parallel transformer language models on unreliable, heterogeneous nodes
- Alpa - Auto parallelization for large-scale neural networks using Jax, XLA, and Ray
Ray + Database
- Balsa Balsa is a learned SQL query optimizer. It tailor optimizes your SQL queries to find the best execution plans for your hardware and engine.
- RaySQL Distributed SQL Query Engine in Python using Ray
- Quokka Open source SQL engine in Python
Ray + X (integration)
- Ray MCP Server – Bridge AI assistants to Ray clusters; manage clusters, submit jobs, and monitor resources through a Model Context Protocol interface.
- prefect-ray Prefect integrations with Ray
- xgboost_ray Distributed XGBoost on Ray
- Ray-DeepSpeed-Inference Run deepspeed on ray serve
Ray-Project
- SkyPilot a framework for easily running machine learning workloads on any cloud through a unified interface
- Exoshuffle-CloudSort the winning entry of the 2022 CloudSort Benchmark in the Indy category.
distributed computing
- Fugue a unified interface for distributed computing that lets users execute Python, pandas, and SQL code on Ray without rewrites.
- Daft is a fast, Pythonic and scalable open-source dataframe library built for Python and Machine Learning workloads.
- Flower(flwr) is a framework for building federated learning systems. Uses Ray for scaling out experiments from desktop, single GPU rack, or multi-node GPU cluster.
- Modin: Scale your pandas workflows by changing one line of code. Uses Ray for transparently scaling out to multiple nodes.
- Volcano is a batch system built on Kubernetes. It provides a suite of mechanisms that are commonly required by many classes of batch & elastic workloads.
Ray AIR
- Ray on Azure ML Turning AML compute into Ray cluster
Cloud Deployment
- Ray on AWS - Official guide for launching Ray clusters on AWS with CloudWatch monitoring
- Ray on GCP - Official guide for launching Ray clusters on Google Cloud Platform
- Ray on Azure - Official guide for launching Ray clusters on Microsoft Azure
Misc
- AutoGluon AutoML for Image, Text, and Tabular Data
- Aws-samples Ray on Amazon SageMaker/EC2/EKS/EMR
- KubeRay A toolkit to run Ray applications on Kubernetes
- ray-educational-materials This is suite of the hands-on training materials that shows how to scale CV, NLP, time-series forecasting workloads with Ray.
- Metaflow-Ray An extension for Metaflow that enables seamless integration with Ray
Videos
Anyscale Academy & Official Tutorials
- Anyscale YouTube Channel - Official YouTube channel with Ray tutorials, conference talks, and educational content
- Anyscale Academy - Ray tutorials from Anyscale with accompanying videos on YouTube
- Ray Crash Course - Introductory online class with video on Anyscale YouTube
- Reinforcement Learning with Ray RLlib - Complete tutorial with video
Conference Talks
- Ray Summit 2024 - Annual Ray conference with recorded sessions on YouTube (Sep 30 - Oct 2, 2024)
- Ray Summit 2025 - Upcoming conference (Nov 3-5, 2025, San Francisco)
RLlib
- Deep reinforcement learning at Riot Games by Ben Kasper - reinforcement learning for game development in production
Papers
This section contains papers focused on Ray (e.g. RAY-based library whitepapers, research on RAY, etc). Papers implemented in RAY are listed in the Models/Projects section.
Foundational Papers
- Ray: A Distributed Framework for Emerging AI Applications (OSDI 2018) - The foundational paper presenting Ray's unified interface for task-parallel and actor-based computations. Demonstrates scaling beyond 1.8 million tasks per second.
- Ray on arXiv - arXiv version of the foundational Ray paper
Tutorials and Blog Posts
2024-2025
- Ray: Your Gateway to Scalable AI and Machine Learning Applications - Analytics Vidhya (March 2025) - Comprehensive guide to Ray's architecture and capabilities with practical project implementation
- RAY: A Powerful Distributed Computing Framework for ML/AI - Spheron Network (June 2024) - Covers Ray's capabilities for scaling models and distributed computing
- The Modern AI Stack: Ray - Medium (September 2024) - How Ray fits into the modern AI infrastructure
- Understanding Iterations in Ray RLlib - tecRacer (February 2024) - Deep dive into RLlib's learning iterations
- Ray Summit 2024: Breaking Through the AI Complexity Wall - Anyscale (2024) - Highlights from Ray Summit 2024, orchestrating 1M+ clusters per month
- How Ray Helps Power ChatGPT - The New Stack - How OpenAI uses Ray for ChatGPT training coordination
Earlier Resources
- Programming in Ray: Tips for first-time users - Berkeley RISE Lab
- Hacker News Discussion - Community discussion about Ray
- Load PyTorch Models 340 Times Faster with Ray - IBM
- Writing Your First Distributed Python Application with Ray - Anyscale official tutorial
books
- Learning Ray Learning Ray - Flexible Distributed Python for Machine Learning
course
- RL course Applied Reinforcement Learning with RLlib
- MLops course MLops course
cheatsheet
Tags: ai distribution
Last modified 20 January 2026