Anyscale

Anyscale is the commercial platform built on Ray, the open-source distributed computing framework created at UC Berkeley's RISELab in 2016. The company was founded in 2019 by Robert Nishihara, Philipp Moritz, Ion Stoica, and Michael I. Jordan — Stoica also co-founded Databricks and was one of the original developers of Apache Spark, and Jordan is among the most cited AI researchers in computer science. Ray runs at more than 10,000 organizations including OpenAI, which used it to train GPT-4, and Anyscale provides the managed cluster layer on top with autoscaling, fault tolerance, and RayTurbo optimizations across AWS, GCP, and Azure.

Key Capabilities:

• Managed Ray clusters on AWS, GCP, and Azure with no infrastructure management

• RayTurbo enterprise-grade Ray with additional reliability and fault-tolerance improvements

• Ray Core for general distributed task and actor-based computing in Python

• Ray Train for distributed ML training across PyTorch, TensorFlow, and Hugging Face

• Ray Tune for distributed hyperparameter optimization

• Ray Serve for scalable multi-model deployment and inference serving

• Ray RLlib for distributed reinforcement learning

• Ray Data for distributed data ingestion and preprocessing in ML pipelines

• Laptop-to-cluster scaling with minimal code changes via Python decorators

• Autoscaling with automatic shutdown of idle resources for cost optimization

• Multi-GPU, CPU, and accelerator support across hardware types

• Cloud-based IDEs including VSCode and Jupyter for remote development

• Dependency management and fault-tolerant cluster configuration

See Anyscale Pricing Details →