What is this?
Ray is an open-source framework for scaling AI and Python applications: a core distributed runtime plus a set of AI libraries that simplify ML compute.
Ray is an open-source framework for scaling AI and Python applications: a core distributed runtime plus a set of AI libraries that simplify ML compute.Is it one library or a whole suite of them?A suite. On top of the Ray Core runtime sit five AI libraries: Data, Train, Tune, RLlib, and Serve, each targeting one ML workload.
A suite. On top of the Ray Core runtime sit five AI libraries: Data, Train, Tune, RLlib, and Serve, each targeting one ML workload.How do I add Ray to a project?Install it from PyPI with pip install ray; nightly wheels are available via the installation page.
Install it from PyPI with pip install ray; nightly wheels are available via the installation page.Is it only for ML work, or any Python code?Any of it, really: Ray is general-purpose and built to run any kind of workload performantly, though the bundled libraries zero in on ML compute.
Any of it, really: Ray is general-purpose and built to run any kind of workload performantly, though the bundled libraries zero in on ML compute.Where can my Ray workloads run?Anywhere: Ray runs on any machine, cluster, cloud provider, and Kubernetes, with a growing ecosystem of community integrations.
Anywhere: Ray runs on any machine, cluster, cloud provider, and Kubernetes, with a growing ecosystem of community integrations.What does the core runtime actually give me?Three abstractions: Tasks (stateless functions executed in the cluster), Actors (stateful worker processes created in the cluster), and Objects (immutable values accessible across the cluster).
Three abstractions: Tasks (stateless functions executed in the cluster), Actors (stateful worker processes created in the cluster), and Objects (immutable values accessible across the cluster).My script runs fine on my laptop — when do I need Ray?When compute intensity outgrows one box: single-node environments like a laptop cannot keep up with compute-heavy ML workloads, and Ray scales the same code from a laptop to a cluster without other infrastructure.
When compute intensity outgrows one box: single-node environments like a laptop cannot keep up with compute-heavy ML workloads, and Ray scales the same code from a laptop to a cluster without other infrastructure.Can I watch what a running cluster is doing?Yes — monitor Ray apps and clusters with the built-in Ray Dashboard, and debug them with the Ray Distributed Debugger.
Yes — monitor Ray apps and clusters with the built-in Ray Dashboard, and debug them with the Ray Distributed Debugger.What are similar projects?wandb, nni, higgsfield, serve, horovod, petals, Made-With-ML, jeeves
wandb, nni, higgsfield, serve, horovod, petals, Made-With-ML, jeevesWhich topics is it in?scaling, hyperparameter-optimization, distributed-computing, parallel-computing, parallel, llm-serving, distributed, distributed-training, model-serving, mlops, optimization, tensorflow
scaling, hyperparameter-optimization, distributed-computing, parallel-computing, parallel, llm-serving, distributed, distributed-training, model-serving, mlops, optimization, tensorflow