# ray-project/ray

Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.

Repository: https://github.com/ray-project/ray
Canonical: https://ross.abutalabs.com/products/ray
Homepage: https://ray.io
Language: Python
License: Apache-2.0
License Family: permissive
Topics: ray, distributed, parallel, machine-learning, reinforcement-learning, deep-learning, python, rllib, hyperparameter-search, optimization, data-science, hyperparameter-optimization, serving, deployment, pytorch, tensorflow, llm-serving, large-language-models, llm, llm-inference
Last push: 2026-08-26T18:17:51+00:00

## Health v2 (maintenance only)
Score: 99/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 99, longevity 100
- inputs: {"age_days": 3599, "days_push": 7, "days_rel": 10, "gap_med": 18, "n_releases_24m": 36}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 43614, forks 7966 (observed 2026-08-28T04:12:09.092487+00:00)

## What it is
Ray is a unified open-source framework for scaling AI and Python applications, consisting of a core distributed runtime (tasks, actors, objects) plus AI libraries for data processing, distributed training, hyperparameter tuning, reinforcement learning, and model serving. It scales workloads from a laptop to clusters with thousands of GPUs across any cloud or Kubernetes.

## Use cases
- distribute python workloads across a cluster
- run distributed model training on multiple gpus
- tune hyperparameters at scale
- serve llm models with autoscaling
- process large datasets for machine learning pipelines
- train reinforcement learning agents
- run parallel simulations or backtesting in python

## When to choose
- you need to scale Python or ML workloads from a single machine to large clusters
- you want distributed training, tuning, serving, and data processing under one framework
- you need fine-grained heterogeneous GPU/CPU scheduling for AI workloads
- you want a Python-native alternative to raw Spark or Kubernetes job orchestration for ML

## When to avoid
- you only need simple single-machine parallelism where multiprocessing suffices
- your workloads are non-ML ETL better served by Spark or a data warehouse
- you need a lightweight setup without cluster management overhead
- you require strict real-time or low-latency guarantees unsuited to distributed scheduling

## Facets
- artifact type: framework
- maturity: stable
- function: machine-learning, deep-learning, llm-inference, llm-training, reinforcement-learning, etl, streaming, gpu-computing, concurrency, deployment, benchmarking, monitoring
- domain: machine-learning, deep-learning, large-language-models, reinforcement-learning, microservices, cloud-computing, artificial-intelligence
- platform: python, windows, cloud
- tags: distributed-computing, hyperparameter-tuning, model-serving, distributed-training, parallel-computing, cluster-computing, rllib, ray-serve, ray-tune, ray-data, data-engineering, linux, macos, kubernetes, docker, gpu

## Member repositories
- ray-project/ray (main) score 99

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:12:09.092487+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T16:22:21.218363+00:00, confidence not recorded.
  - readme: https://github.com/ray-project/ray (fetched 2026-08-28T04:12:09.092487+00:00, sha 016811dbca46)
  - homepage: https://ray.io (fetched 2026-08-29T07:46:12.478422+00:00, sha a250d8990b7a)
- Data as of 2026-08-30T08:39:29.467469+00:00.