Implementation:SeldonIO Seldon core Seldon Model Unload

Property	Value
Implementation Name	Seldon_Model_Unload
Type	External Tool Doc
Overview	Concrete CLI tool for unloading models and managing model lifecycle in Seldon Core 2.
Implements Principle	SeldonIO_Seldon_core_Model_Lifecycle_Management
Workflow	Model_Deployment
Domains	MLOps, Kubernetes
Source	docs-gb/cli/seldon_model_unload.md:L1-20
External Dependencies	seldon CLI, kubectl
Last Updated	2026-02-13 00:00 GMT

Description

The seldon model unload command removes a model from the Seldon Core 2 scheduler, triggering the inference server to unload the model from memory and free associated resources. This is the primary tool for decommissioning models and managing server capacity. For rolling updates (deploying a new version of an existing model), the seldon model load command is used instead with an updated Model CRD, and the scheduler handles the version transition automatically.

The equivalent operation via kubectl is kubectl delete model <name>, which removes the Model custom resource and triggers the same unload process through the Kubernetes operator.

Code Reference

Source: docs-gb/cli/seldon_model_unload.md:L1-20

CLI Signature:

seldon model unload <modelName> [--scheduler-host string]

Alternatives:

# Unload via kubectl
kubectl delete model <name>

# Rolling update (load new version, scheduler transitions automatically)
seldon model load -f <updated.yaml>

Key Parameters

Parameter	Type	Default	Description
modelName	string (positional)	(required)	Name of the model to unload from the inference server
-f / --file-path	string	(none)	Model manifest file path (alternative to positional name for unload-by-file)
--scheduler-host	string	"0.0.0.0:9004"	Address of the Seldon scheduler gRPC endpoint
--force	boolean	false	Force the unload operation in control plane mode
-h / --help	boolean	false	Display help information for the command

I/O Contract

Inputs

Input	Type	Description
Model name	string	The name of a currently deployed Seldon Core 2 model to unload
Updated model YAML (for rolling update)	File path	A Model CRD with updated storageUri pointing to the new model version

Outputs

Output	Type	Description
Unload confirmation	JSON	Empty JSON {} on successful unload from the scheduler
Traffic shift (rolling update)	Scheduler event	Traffic gradually shifted to the new model version, old version drained and unloaded

Usage Examples

Basic Model Unload

# Unload a model by name
seldon model unload iris

Unload via kubectl

# Delete the Model custom resource
kubectl delete model iris

Rolling Update to New Version

# Update the storageUri in the YAML to point to v2 artifact
# samples/models/sklearn1-v2.yaml:
#   storageUri: "gs://seldon-models/mlserver/iris/v2"

# Resubmit the updated manifest
seldon model load -f samples/models/sklearn1-v2.yaml

# Wait for the new version to be ready
seldon model status iris -w ModelAvailable

# Verify inference works with the new version
seldon model infer iris \
  '{"inputs": [{"name": "predict", "shape": [1, 4], "datatype": "FP32", "data": [[5.1, 3.5, 1.4, 0.2]]}]}'

Full Lifecycle Workflow

# Step 1: Deploy initial model
seldon model load -f samples/models/sklearn1.yaml
seldon model status iris -w ModelAvailable

# Step 2: Run inference
seldon model infer iris \
  '{"inputs": [{"name": "predict", "shape": [1, 4], "datatype": "FP32", "data": [[5.1, 3.5, 1.4, 0.2]]}]}'

# Step 3: Deploy updated model (rolling update)
seldon model load -f samples/models/sklearn1-v2.yaml
seldon model status iris -w ModelAvailable

# Step 4: Verify new version
seldon model infer iris \
  '{"inputs": [{"name": "predict", "shape": [1, 4], "datatype": "FP32", "data": [[5.1, 3.5, 1.4, 0.2]]}]}'

# Step 5: Decommission when no longer needed
seldon model unload iris

Knowledge Sources

• Repository: https://github.com/SeldonIO/seldon-core
• Documentation: https://docs.seldon.io/projects/seldon-core/en/v2/

Related Pages

• SeldonIO_Seldon_core_Seldon_Model_Unload implements SeldonIO_Seldon_core_Model_Lifecycle_Management
• SeldonIO_Seldon_core_Seldon_Model_Load is reused for rolling updates by SeldonIO_Seldon_core_Seldon_Model_Unload
• SeldonIO_Seldon_core_Seldon_Model_Status verifies updates after SeldonIO_Seldon_core_Seldon_Model_Unload
• SeldonIO_Seldon_core_Seldon_Model_Infer validates new versions after SeldonIO_Seldon_core_Seldon_Model_Unload
• Environment:SeldonIO_Seldon_core_Kubernetes_Cluster_Environment
• Heuristic:SeldonIO_Seldon_core_Model_Load_Timeout_Tip