Deploy Serialized Models

Last updated 10th January, 2020.

Objective

Deploying models is the main feature of the ML Serving. This guide provides step by step instructions on how to deploy a serialized model.

Requirements

• Having previously created a ML Serving namespace. Steps are described in section Initializing namespace
• Having previously exported your machine learning model into a supported format. Steps to export TensorFlow models are described here, steps to export Scikit-Learn models are described here.

Instructions

Step 1 - Upload the serialized model into the public cloud object storage

You need to upload your serialized model into the object storage container which is linked with your namespace. On the main page of your wanted namespace you can click on the name of your linked container under the Information > Container Object Storage section.

This step is just a redirection on Public Cloud > Storage > Object Storage > <your container>.

You can then click the Add object button and select your exported model from your local file storage.

After clicking the Import button you should see your exported model in the list of containers object.

Step 2 - Reach model page

Inside your ML Serving namespace page there is a dedicated tab for managing models : Models.

You can start the deployment of a new model by clicking the Deploy Model button.

Step 3 - Select the custom model

In this guide, we explain how to deploy a serialized model also called custom model. Just select custom model and click Next.

Step 4 - Select your exported file

Here you will click on file and search on the dropdown list your model and click Next.

Step 5 - Choose a name for your model

That name identifies your model among others on your namespace.

After you filled your wanted name, click the Next button.

Step 6 - Choose your auto-scaling options

A model is composed of one or several running instance(s). These instances are automaticaly scaled by the ML Serving depending on the input load of your model. That step allows you to configure the minimum and maximum number of instances that you want to run.

Step 7 - Choose your instances configuration

Each model instance is related to a CPU & RAM flavor. You can choose the wanted flavor among a list of existing ones.

Step 8 - Wait for model deployment

The ML Serving will sequentially perform the following tasks :

• The building of the model : you see the build status as Building.
• The deployment task : you see the API Status as Pending.

When everything is up and running you see the build status as Deployed and the api status as Running. The URL where to reach your model is also displayed so you can start requesting it.

Step 9 - Check the model web interface

When you follow the model URL in your web browser, you should arrive on the web page that describes the available API.

This page should look like this :

There are two endpoints available on that kind of model :

• An endpoint to describe the model's inputs and outputs
• An endpoint to query the model

The web interface allows you to interactively execute queries on the different endpoints of your model if you want to test quickly.

Going Further

• You can check the official documentation of the open source product ML Serving about Models
• You can check the supported compatibilities for custom models

---

---

These guides might also interest you...