Implementation:SeldonIO Seldon core Seldon Model Load For Pipeline

Field	Value
Implementation Name	Seldon Model Load For Pipeline
Type	External Tool Doc
Overview	Concrete CLI tool for deploying multiple component models as pipeline prerequisites in Seldon Core 2.
Related Principle	SeldonIO_Seldon_core_Component_Model_Deployment
Source	samples/models/tfsimple1.yaml:L1-9, samples/pipeline-examples.md:L42-55
Domains	MLOps, Kubernetes
External Dependencies	seldon CLI, kubectl, Seldon scheduler, GCS storage
Knowledge Sources	Repo (https://github.com/SeldonIO/seldon-core)
Last Updated	2026-02-13 00:00 GMT

Description

This implementation covers the concrete steps for deploying individual models that will serve as building blocks for Seldon Core 2 Pipelines. Each model is defined as a Kubernetes custom resource of kind Model under the mlops.seldon.io/v1alpha1 API group, then loaded onto the Seldon scheduler using the seldon model load CLI command.

Code Reference

Model CRD YAML (tfsimple1)

apiVersion: mlops.seldon.io/v1alpha1
kind: Model
metadata:
  name: tfsimple1
spec:
  storageUri: "gs://seldon-models/triton/simple"
  requirements:
  - tensorflow
  memory: 100Ki

Source: samples/models/tfsimple1.yaml:L1-9

Model CRD YAML (tfsimple2)

apiVersion: mlops.seldon.io/v1alpha1
kind: Model
metadata:
  name: tfsimple2
spec:
  storageUri: "gs://seldon-models/triton/simple"
  requirements:
  - tensorflow
  memory: 100Ki

Source: samples/models/tfsimple2.yaml:L1-9

Key Fields

Field	Description	Example
metadata.name	Unique model identifier referenced by pipelines	tfsimple1
spec.storageUri	URI to model artifacts in cloud storage	gs://seldon-models/triton/simple
spec.requirements	List of framework requirements for server selection	[tensorflow]
spec.memory	Memory allocation for the model	100Ki

I/O Contract

Inputs

• Model CRD YAML files: One YAML manifest per component model, each specifying storageUri, requirements, and memory.

Outputs

• Multiple models loaded and available on inference servers: Each model transitions through states until reaching ModelAvailable, at which point it can serve inference requests and be referenced by pipelines.

Usage Examples

Loading Models via CLI

# Load the first component model
seldon model load -f ./models/tfsimple1.yaml

# Load the second component model
seldon model load -f ./models/tfsimple2.yaml

Waiting for Models to Become Available

# Wait for each model to reach ModelAvailable state
seldon model status tfsimple1 -w ModelAvailable | jq -M .
seldon model status tfsimple2 -w ModelAvailable | jq -M .

Alternative: Loading via kubectl

kubectl apply -f ./models/tfsimple1.yaml
kubectl apply -f ./models/tfsimple2.yaml

CLI Signature

seldon model load -f <model.yaml> [--scheduler-host string] [--force] [-v]

Flag	Description	Default
-f, --file-path	Model manifest file (YAML)	(required)
--scheduler-host	Seldon scheduler host	0.0.0.0:9004
--force	Force control plane mode	false
-v, --verbose	Verbose output	false

Related Pages

• SeldonIO_Seldon_core_Component_Model_Deployment - implements - Principle describing why models must be independently deployed before pipeline creation.
• SeldonIO_Seldon_core_Seldon_Pipeline_CRD - next step - Defining the pipeline topology that references these loaded models.
• SeldonIO_Seldon_core_Seldon_Pipeline_Load - next step - Loading the pipeline after all models are available.