Implementation:Kserve Kserve TrainedModel CRD Spec

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

Overview

Concrete Go type definitions for the TrainedModel CRD that registers individual model artifacts with a shared InferenceService.

Description

The TrainedModel CRD is defined in pkg/apis/serving/v1alpha1/trained_model.go. The spec contains an InferenceService reference (parent ISVC name), a Model field with StorageURI, Framework, and Memory. The webhook validates naming, storage URI scheme, and memory constraints.

Usage

Create TrainedModel resources pointing to a running MMS InferenceService. Each TrainedModel gets its own prediction endpoint under the shared ISVC host.

Code Reference

Source Location

• Repository: kserve
• File: pkg/apis/serving/v1alpha1/trained_model.go, Lines 33-81

Signature

// TrainedModel registers a model with a shared InferenceService
type TrainedModel struct {
    metav1.TypeMeta   `json:",inline"`
    metav1.ObjectMeta `json:"metadata,omitempty"`
    Spec              TrainedModelSpec   `json:"spec,omitempty"`
    Status            TrainedModelStatus `json:"status,omitempty"`
}

// TrainedModelSpec defines the model to register
type TrainedModelSpec struct {
    InferenceService string    `json:"inferenceService"`
    Model            ModelSpec `json:"model"`
}

// ModelSpec defines the model artifact
type ModelSpec struct {
    StorageURI string            `json:"storageUri"`
    Framework  string            `json:"framework"`
    Memory     resource.Quantity `json:"memory"`
}

// TotalRequestedMemory sums memory of all TrainedModels for a parent ISVC
func TotalRequestedMemory(models []TrainedModel) resource.Quantity

Import

import servingv1alpha1 "github.com/kserve/kserve/pkg/apis/serving/v1alpha1"

I/O Contract

Inputs

Name	Type	Required	Description
spec.inferenceService	string	Yes	Name of parent MMS InferenceService
spec.model.storageUri	string	Yes	Model artifact URI (s3://, gs://, https://)
spec.model.framework	string	Yes	ML framework (sklearn, xgboost, pytorch, etc.)
spec.model.memory	resource.Quantity	Yes	Estimated model memory usage

Outputs

Name	Type	Description
status.conditions	[]Condition	InferenceServiceReady, IsMMSPredictor, MemoryResourceAvailable
status.url	URL	Per-model prediction endpoint URL
ConfigMap entry	JSON	Model spec written to modelconfig-<isvc>-0

Usage Examples

Register Two Models

apiVersion: serving.kserve.io/v1alpha1
kind: TrainedModel
metadata:
  name: model1-sklearn
spec:
  inferenceService: sklearn-iris-example
  model:
    storageUri: "gs://kfserving-examples/models/sklearn/1.0/model"
    framework: sklearn
    memory: 256Mi
---
apiVersion: serving.kserve.io/v1alpha1
kind: TrainedModel
metadata:
  name: model2-sklearn
spec:
  inferenceService: sklearn-iris-example
  model:
    storageUri: "gs://kfserving-examples/models/sklearn/1.0/model-2"
    framework: sklearn
    memory: 256Mi

kubectl apply -f trainedmodels.yaml
kubectl get trainedmodels
# NAME              URL   READY
# model1-sklearn          True
# model2-sklearn          True

Related Pages

Implements Principle

• Principle:Kserve_Kserve_TrainedModel_Registration