Ray

Ray

Ray is a unified framework for scaling Python and AI/ML workloads across a cluster. It provides low-level distributed primitives (tasks, actors) and a rich library ecosystem — Ray Train, Ray Serve, Ray Tune, RLlib, and Ray Data — that handle the most common distributed AI use cases out of the box.

Key Features

• Ray Train: distributed training for PyTorch, TensorFlow, XGBoost, LightGBM with fault tolerance
• Ray Serve: production model serving with request batching, model composition, and autoscaling
• Ray Tune: distributed hyperparameter search with Bayesian optimization and PBT
• Ray Data: scalable data preprocessing pipelines for ML training
• RLlib: industry-grade reinforcement learning library
• Scales from laptop to cluster: same code, no changes

Quick Start

import ray

ray.init()

@ray.remote
def process_batch(batch):
    return [preprocess(item) for item in batch]

# Fan out 1000 tasks in parallel
futures = [process_batch.remote(batch) for batch in batches]
results = ray.get(futures)

Install via ai-supply

npx ai-supply add ray-distributed-computing

Curated mirror of the open-source Ray (Apache-2.0). Get it from the source.