Ray

The product is an open source Python-native framework that orchestrates infrastructure to run distributed AI and ML workloads across heterogeneous CPUs and GPUs at scale, addressing slow production, underutilized accelerators and rising costs. It is for developers, data teams and operations teams in businesses building AI platforms. It is delivered as an open source compute engine with managed cloud options that supports any workload, data type and accelerator with fine-grained scaling from a laptop to thousands of GPUs.

Key features

• Parallel Python code scaling
• Multi-modal data processing
• Distributed model training
• Model serving with independent scaling
• Batch inference on heterogeneous compute
• Reinforcement learning workflows
• GenAI and RAG workflows
• LLM inference and fine-tuning
• Core primitives for distributed apps
• Ray libraries for end-to-end AI