Ray

An open-source distributed computing framework for scaling Python AI and ML workloads from a single machine to a large cluster without rewriting code. Ray's core model lets any Python function run as a distributed task and any Python class run as a distributed stateful actor, making parallel and distributed execution almost as easy as regular Python. Ray Tune provides distributed hyperparameter optimization across hundreds of parallel training jobs. Ray Train scales model training in PyTorch and TensorFlow across multiple GPUs and machines. Ray Serve deploys ML models as production online services with batching, autoscaling, and model composition support. Ray Data handles large-scale data preprocessing in parallel pipelines. Used by every major AI company and research lab for scaling LLM training, reinforcement learning environments, and inference workloads. Open source under Apache 2.0 on GitHub; managed cloud version is Anyscale. Used by companies including OpenAI, Anthropic, and Uber.

Browse tasks in this category · Category overview