Labels and Annotations

Reference of labels and annotations used by OME.

This document serves as a reference of the various labels and annotations used throughout OME.

Annotations

InferenceService Annotations

These annotations are used to configure InferenceService behavior:

Annotation	Description
`ome.io/enable-tag-routing`	Enables tag-based routing for the InferenceService
`ome.io/autoscalerClass`	Specifies the autoscaler class to use. Valid values: `hpa`, `keda`, `external`
`ome.io/metrics`	Defines the scaling metric type. Valid values: `cpu`, `memory`
`ome.io/targetUtilizationPercentage`	Sets the target utilization percentage for autoscaling
`ome.io/deprecation-warning`	Displays deprecation warnings for legacy configurations
`ome.io/enable-metric-aggregation`	Enables metric aggregation for the InferenceService
`ome.io/enable-prometheus-scraping`	Enables Prometheus scraping for metrics collection
`ome.io/volcano-queue`	Specifies the Volcano queue name for job scheduling

Model and Runtime Annotations

Annotation	Description
`ome.io/inject-model-init`	Enables injection of model initialization containers
`ome.io/inject-fine-tuned-adapter`	Enables injection of fine-tuned adapter containers
`ome.io/inject-serving-sidecar`	Enables injection of serving sidecar containers
`ome.io/fine-tuned-weight-ft-strategy`	Specifies the fine-tuning strategy for weights
`ome.io/base-model-name`	Specifies the base model name
`ome.io/base-model-vendor`	Specifies the base model vendor
`ome.io/serving-runtime`	Specifies the serving runtime to use
`ome.io/base-model-format`	Specifies the base model format
`ome.io/base-model-format-version`	Specifies the base model format version
`ome.io/fine-tuned-serving-with-merged-weights`	Enables fine-tuned serving with merged weights

Model Security Annotations

These annotations control model encryption and decryption:

Annotation	Description
`ome.io/base-model-decryption-key-name`	Specifies the decryption key name for the base model
`ome.io/base-model-decryption-secret-name`	Specifies the secret name containing decryption credentials
`ome.io/disable-model-decryption`	Disables model decryption

Service Configuration Annotations

Annotation	Description
`ome.io/service-type`	Specifies the Kubernetes service type
`ome.io/load-balancer-ip`	Sets the load balancer IP address

RDMA Annotations

Annotation	Description
`rdma.ome.io/auto-inject`	Enables automatic RDMA injection
`rdma.ome.io/profile`	Specifies the RDMA profile to use
`rdma.ome.io/container-name`	Specifies the container name for RDMA configuration

Knative Annotations

Annotation	Description
`autoscaling.knative.dev/min-scale`	Sets the minimum number of replicas
`autoscaling.knative.dev/max-scale`	Sets the maximum number of replicas
`serving.knative.dev/rollout-duration`	Specifies the rollout duration
`serving.knative.openshift.io/enablePassthrough`	Enables passthrough on OpenShift

Runtime Revision and Pinning Annotations

These annotations drive runtime revision pinning, which lets an InferenceService pin to a content-addressed snapshot of its ServingRuntime instead of always tracking the live runtime.

Annotation	Description
`ome.io/runtime-sync`	Set/bump to a new value on a pinned InferenceService (`spec.runtime.autoSync: false`) to acknowledge runtime drift and advance the pin to a fresh runtime snapshot.
`ome.io/gc-eligible-since`	Set by the controller on an OME-managed ControllerRevision (RFC3339 timestamp) when it first becomes unreferenced and over the retention count; the garbage collector uses it. Not user-set.

Deployment Configuration Annotations

Annotation	Description
`ome.io/deploymentMode`	Selects the deployment strategy (e.g. `RawDeployment`, `MultiNode`, `Serverless`, `PDDisaggregated`).
`ome.io/dedicated-ai-cluster`	Associates the InferenceService with a dedicated AI cluster.
`ome.io/entrypoint-component`	Identifies the entrypoint component of a multi-component InferenceService.
`ome.io/accelerator-class`	Selects the accelerator class used for runtime matching and scheduling.

Ingress Annotations

See Ingress administration for usage details.

Annotation	Description
`ome.io/ingress-domain`	Overrides the ingress domain for the InferenceService.
`ome.io/ingress-domain-template`	Template used to construct the ingress domain.
`ome.io/ingress-additional-domains`	Additional domains to expose the InferenceService on.
`ome.io/ingress-url-scheme`	URL scheme (`http`/`https`) for generated ingress URLs.
`ome.io/ingress-path-template`	Template used to construct the ingress path.
`ome.io/ingress-disable-istio-virtualhost`	Disables creation of the Istio VirtualHost for the InferenceService.
`ome.io/ingress-disable-creation`	Disables ingress creation entirely for the InferenceService.

Observability Annotations

Annotation	Description
`prometheus.ome.io/port`	Container port Prometheus should scrape for metrics.
`prometheus.ome.io/path`	HTTP path Prometheus should scrape for metrics.

Model Management Annotations

Annotation	Description
`ome.oracle.com/skip-config-parsing`	Set on a BaseModel/ClusterBaseModel to skip automatic model-config parsing by the model agent.

Labels

Model and Runtime Labels

Label	Description
`base-model-name`	Base model name label
`base-model-size`	Base model size label
`base-model-type`	Base model type label
`base-model-vendor`	Base model vendor label
`fine-tuned-serving`	Fine-tuned serving label
`fine-tuned-serving-with-merged-weights`	Fine-tuned serving with merged weights label
`serving-runtime`	Serving runtime label
`fine-tuned-weight-ft-strategy`	Fine-tuning strategy label

Scheduling Labels

Label	Description
`ray.io/scheduler-name`	Ray scheduler name
`ray.io/priority-class-name`	Ray priority class name
`raycluster/unavailable-since`	Ray cluster unavailable timestamp
`volcano.sh/queue-name`	Volcano queue name
`volcano.sh/job-name`	Volcano job name

Kueue Labels

Label	Description
`kueue.x-k8s.io/queue-name`	Kueue queue name
`kueue.x-k8s.io/priority-class`	Kueue workload priority class
`kueue-enabled`	Enables Kueue for the resource

Model Agent Labels

Label	Description
`node.kubernetes.io/instance-type`	Node instance shape
`models.ome/{uid}`	Model label with UID
`models.ome.io/target-instance-shapes`	Target instance shapes for models
`models.ome/basemodel-status`	Base model status

Component Labels

Label	Description
`component`	KService component label
`endpoint`	KService endpoint label
`ome.io/inferenceservice`	InferenceService label for TrainedModel
`ome.io/inferenceservice`	InferenceService pod label

Network Visibility Labels

Label	Description
`networking.ome.io/visibility`	Network visibility configuration
`networking.knative.dev/visibility`	Knative network visibility
`sidecar.istio.io/inject`	Istio sidecar injection

Runtime Revision Labels

Applied by the controller to the OME-managed ControllerRevisions used for runtime pinning. Not user-set.

Label	Description
`ome.io/runtime-of`	Name of the source ServingRuntime/ClusterServingRuntime the revision snapshots.
`ome.io/runtime-of-kind`	Kind of the source runtime (`ServingRuntime` or `ClusterServingRuntime`).
`ome.io/runtime-of-namespace`	Namespace of the source runtime (empty for cluster-scoped).
`ome.io/revision-hash`	Content hash of the snapshotted runtime spec; used to find-or-create revisions.
`ome.io/created-by`	Marks the revision as OME-created (value `ome-controller`); gates the immutability webhook and GC.

Model Management Labels

Label	Description
`models.ome.io/category`	Category of the model.
`models.ome.io/node-name`	Node the model artifact is managed on (set by the model agent).
`models.ome.io/managed-by`	Identifies the manager of the model artifact.
`models.ome/reserve-model-artifact`	Marks a model artifact to be reserved (retained, not reclaimed).

Special Values

Autoscaler Classes

hpa: Horizontal Pod Autoscaler
keda: Kubernetes Event-driven Autoscaling
external: External autoscaler

Scale Metrics

cpu: CPU utilization
memory: Memory utilization
concurrency: Request concurrency (Knative)
rps: Requests per second (Knative)

Priority Classes

volcano-scheduling-high-priority: High priority for Volcano scheduling
kueue-scheduling-high-priority: High priority for Kueue workload scheduling

Feedback

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.

Last modified July 1, 2026: [Docs] update docs for new features (#648) (8c31d39)