OME API
Generated API reference documentation for ome.io/v1beta1.
Generated API reference documentation for ome.io/v1beta1.
Reference of labels and annotations used by OME.
When runtime auto-selection rejects a runtime because the InferenceService names an AcceleratorClass the runtime does not list, and what happens instead when the runtime was named explicitly.
Reference for the storage backends a BenchmarkJob’s spec.outputLocation can target — the storageUri format and the parameters keys each backend accepts.
Reference for metric-gated canary promotion: where analysis metrics come from, how interval, initialDelay, and failureLimit shape the verdict, which template variables a query may use, and what Hold, Rollback, and RollbackOnStall do when Prometheus cannot be read.
Reference for spec.rollout.groups[].canary steps — what capacity and traffic each control, how a step’s gate (immediate, timed, manual, analysis) is resolved, and what scaleDownDelaySeconds and readyTimeout bound.
When runtime auto-selection rejects a runtime because of a declared deployment mode, and how to declare modes on a runtime’s engineConfig/decoderConfig and on InferenceService components.
How supportedModelFormats[].diffusionPipeline constrains runtime selection to specific diffusers pipelines and their components.
The shared safety contract behind every mutating kubectl-ome command: dry-run modes, confirmation, the guarded patch, the ActionResult output, and exit codes.
How version constraints in a runtime’s supportedModelFormats are matched against a model’s modelFormat and modelFramework versions.
Reference for the ome.io/* annotations that tune backend traffic resilience on an InferenceService — circuit breaking, retries, and connection timeouts — plus the raw pass-through prefixes for Envoy Gateway and Istio.
Was this page helpful?
Glad to hear it! Please tell us how we can improve.
Sorry to hear that. Please tell us how we can improve.