KFServing URI Storage

The kfserving project provides a serverless framework to run many Machine Learning model frameworks on Kubernetes with minimal effort from data scientists.

In order to allow for easier testing, learning, and proof-of-concept (PoC) work, the ability to define a generic URI was added in this PR. This repo provides examples of defining (very minimal) models in various frameworks and their frozen representations, along with the Custom Resource Definition (CRD) to apply them to an existing Kubernetes cluster with kfserving running.

Example

The easiest model example is likely that of Scikit-Learn which most students and professionals are familiar with. In ./sklearn/model/train.py, we specify a sample Iris dataset model to predict flowers,

from sklearn import svm
from sklearn import datasets
import joblib

def train(X, y):
    clf = svm.SVC(gamma='auto')
    clf.fit(X, y)
    return clf

def freeze(clf, path='../frozen'):
    joblib.dump(clf, f'{path}/model.joblib')
    return True

if __name__ == '__main__':
    iris = datasets.load_iris()
    X, y = iris.data, iris.target
    clf = train(X, y)
    freeze(clf)

You can use the frozen model object in ./sklearn/frozen/model.joblib to test out deploying this model in kfserving.

The CRD for that would look like this:

apiVersion: serving.kubeflow.org/v1alpha2
kind: InferenceService
metadata:
  name: sklearn-from-uri
spec:
  default:
    predictor:
      sklearn:
        storageUri: https://github.com/tduffy000/kfserving-uri-examples/blob/master/sklearn/frozen/model.joblib?raw=true

The full example can be found here.