Implementation:Kserve Kserve TrainedModel Full CRD

Knowledge Sources	Kserve_Kserve KServe Docs
Domains	Kubernetes, Multi-Model Serving
Last Updated	2026-02-13 00:00 GMT

Overview

Concrete CustomResourceDefinition for the TrainedModel resource provided by the KServe project.

Description

This file defines the full CRD for the TrainedModel custom resource (short name: tm), which represents individual trained models that can be loaded into a multi-model serving InferenceService. It is a namespaced v1alpha1 resource with spec fields for inferenceService reference and model details (framework, memory, storageUri), plus Knative-style status with conditions, URL, and address fields. This CRD enables multi-model serving by allowing multiple TrainedModel resources to be associated with a single InferenceService.

Usage

Apply this CRD to a Kubernetes cluster to enable multi-model serving workflows. Users create TrainedModel resources that reference an existing InferenceService, specifying the model framework, memory requirements, and storage location. The controller then loads these models into the referenced InferenceService. This follows the principle described in Kserve_Kserve_TrainedModel_Registration.

Code Reference

Source Location

Repository: Kserve_Kserve
File: config/crd/full/serving.kserve.io_trainedmodels.yaml
Lines: 1-111

Signature

apiVersion: apiextensions.k8s.io/v1
kind: CustomResourceDefinition
metadata:
  annotations:
    controller-gen.kubebuilder.io/version: v0.19.0
  name: trainedmodels.serving.kserve.io
spec:
  group: serving.kserve.io
  names:
    kind: TrainedModel
    listKind: TrainedModelList
    plural: trainedmodels
    shortNames:
    - tm
    singular: trainedmodel
  scope: Namespaced
  versions:
  - additionalPrinterColumns:
    - jsonPath: .status.url
      name: URL
      type: string
    - jsonPath: .status.conditions[?(@.type=='Ready')].status
      name: Ready
      type: string
    - jsonPath: .metadata.creationTimestamp
      name: Age
      type: date
    name: v1alpha1
    schema:
      openAPIV3Schema:
        properties:
          spec:
            properties:
              inferenceService:
                type: string
              model:
                properties:
                  framework:
                    type: string
                  memory:
                    x-kubernetes-int-or-string: true
                  storageUri:
                    type: string
                required:
                - framework
                - memory
                - storageUri
            required:
            - inferenceService
            - model

Import

kubectl apply -f config/crd/full/serving.kserve.io_trainedmodels.yaml

I/O Contract

Inputs

Name	Type	Required	Description
spec.inferenceService	string	Yes	Name of the InferenceService to load this model into
spec.model.framework	string	Yes	The ML framework of the model (e.g., sklearn, tensorflow, pytorch)
spec.model.memory	integer or string (Quantity)	Yes	Memory required by the model
spec.model.storageUri	string	Yes	URI pointing to the model artifact storage location

Outputs

Name	Type	Description
TrainedModel CRD	CustomResourceDefinition	Registers the TrainedModel resource type in the Kubernetes API server
status.url	string	The URL at which the trained model is accessible
status.conditions	array	Knative-style conditions indicating readiness and health
status.address	object	Address information including URL, CACerts, and audience

Usage Examples

Apply the CRD

kubectl apply -f config/crd/full/serving.kserve.io_trainedmodels.yaml

Create a TrainedModel resource

apiVersion: serving.kserve.io/v1alpha1
kind: TrainedModel
metadata:
  name: my-sklearn-model
  namespace: default
spec:
  inferenceService: my-inference-service
  model:
    framework: sklearn
    memory: "1Gi"
    storageUri: "gs://my-bucket/sklearn-model"

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment