Skip to content

Deploy Scikit-learn models with InferenceService

This example walks you through how to deploy a scikit-learn model leveraging the v1beta1 version of the InferenceService CRD. Note that, by default the v1beta1 version will expose your model through an API compatible with the existing V1 Dataplane. However, this example will show you how to serve a model through an API compatible with the new V2 Dataplane.


The first step will be to train a sample scikit-learn model. Note that this model will be then saved as model.joblib.

from sklearn import svm
from sklearn import datasets
from joblib import dump

iris = datasets.load_iris()
X, y =,

clf = svm.SVC(gamma='scale'), y)

dump(clf, 'model.joblib')

Testing locally

Once you've got your model serialised model.joblib, we can then use MLServer to spin up a local server. For more details on MLServer, feel free to check the SKLearn example doc.


this step is optional and just meant for testing, feel free to jump straight to deploying with InferenceService.


Firstly, to use MLServer locally, you will first need to install the mlserver package in your local environment, as well as the SKLearn runtime.

pip install mlserver mlserver-sklearn

Model settings

The next step will be providing some model settings so that MLServer knows:

  • The inference runtime to serve your model (i.e. mlserver_sklearn.SKLearnModel)
  • The model's name and version

These can be specified through environment variables or by creating a local model-settings.json file:

  "name": "sklearn-iris",
  "version": "v1.0.0",
  "implementation": "mlserver_sklearn.SKLearnModel"

Note that, when you deploy your model, KServe will already inject some sensible defaults so that it runs out-of-the-box without any further configuration. However, you can still override these defaults by providing a model-settings.json file similar to your local one. You can even provide a set of model-settings.json files to load multiple models.

Serving model locally

With the mlserver package installed locally and a local model-settings.json file, you should now be ready to start our server as:

mlserver start .

Deploy with InferenceService

Lastly, you will use KServe to deploy the trained model. For this, you will just need to use version v1beta1 of the InferenceService CRD and set the protocolVersion field to v2.

apiVersion: ""
kind: "InferenceService"
  name: "sklearn-irisv2"
      protocolVersion: "v2"
      storageUri: "gs://seldon-models/sklearn/mms/lr_model"
apiVersion: ""
kind: "InferenceService"
  name: "sklearn-irisv2"
        name: sklearn
      runtime: kserve-mlserver
      storageUri: "gs://seldon-models/sklearn/mms/lr_model"

Note that this makes the following assumptions:

  • Your model weights (i.e. your model.joblib file) have already been uploaded to a "model repository" (GCS in this example) and can be accessed as gs://seldon-models/sklearn/iris.
  • There is a K8s cluster available, accessible through kubectl.
  • KServe has already been installed in your cluster.
kubectl apply -f ./sklearn.yaml

Testing deployed model

You can now test your deployed model by sending a sample request.

Note that this request needs to follow the V2 Dataplane protocol. You can see an example payload below:

  "inputs": [
      "name": "input-0",
      "shape": [2, 4],
      "datatype": "FP32",
      "data": [
        [6.8, 2.8, 4.8, 1.4],
        [6.0, 3.4, 4.5, 1.6]

Now, assuming that your ingress can be accessed at ${INGRESS_HOST}:${INGRESS_PORT} or you can follow this instruction to find out your ingress IP and port.

you can use curl to send the inference request as:

SERVICE_HOSTNAME=$(kubectl get inferenceservice sklearn-irisv2 -o jsonpath='{.status.url}' | cut -d "/" -f 3)

curl -v \
  -H "Host: ${SERVICE_HOSTNAME}" \
  -H "Content-Type: application/json" \
  -d @./iris-input.json \

Expected Output

  "id": "823248cc-d770-4a51-9606-16803395569c",
  "model_name": "sklearn-irisv2",
  "model_version": "v1.0.0",
  "outputs": [
      "data": [1, 1],
      "datatype": "INT64",
      "name": "predict",
      "parameters": null,
      "shape": [2]
Back to top