Live experiment evidence
Publish scalar history, inspect durable artifacts, and open the same run in Stellar
Examples
Choose a runnable TauGrid, Kueue, Ray, or AKS example by learning goal and prerequisite cost.
Choose an example by the question you want to answer:
| Goal | Example | Interface | Compute | Start here |
|---|---|---|---|---|
| Run repository-first GPU HPO and verify six Tune trials | Ray Tune smoke on AKS | TauGrid-first | One or more NVIDIA GPUs | Run GPU Ray Tune HPO |
| Publish live loss and accuracy evidence, retrieve durable files, and open Stellar | Experiment evidence | TauGrid-first | One NVIDIA GPU | Run live experiment evidence |
| See queue admission and borrowing using CPU-only quota | CPU queueing | Raw KubeRay and Kueue YAML | CPU | Explore CPU queueing |
| Build a complete AKS, Kueue, and Ray baseline | Modular cluster deployment | Terraform, Helm, and kubectl | CPU or A100 GPU | Provision the platform baseline |
Read the interface label
TauGrid-first examples demonstrate the researcher interface this site recommends and remains the preferred researcher workflow. Raw KubeRay and Kueue examples expose the Kubernetes objects beneath TauGrid and are useful for platform education or debugging.
The anyscale-comparison directory serves as a vendor-comparison reference
with known external-resource defaults; it lives outside this table and the
supported TauGrid workflow set.
CPU queueing with Kueue and Ray
Observe Kueue admission and quota behavior with raw RayJobs and no GPU quota.
Run GPU Ray Tune HPO on AKS
Provision GPU-capable AKS, enable TauGrid, and run the six-trial Ray Tune example through a repository workspace.
Provision a GPU-enabled TauGrid AKS environment
Build AKS, GPU capacity, TauGrid, and Portal with the repository Terraform root.