Deployment view

The Peltarion deployment solution gives you the:

  • Means to quickly test out model prototypes all the way directly in your services.

  • Stability and scalability you need for a system that will be deployed for longer periods of time with a reliable model for server-to-server integration.

Note: You cannot create a deployment using an archived experiment.

Create standard deployment

Use Standard for classification and regression models.

  1. Click New deployment to create a new deployment.

  2. Select Standard.

  3. Select the Experiment and Checkpoint you want to deploy.

  4. Click Create.

As you do in the car damage tutorial.

Create similarity search deployment

Use Similarity search for image or text similarity search models.

  1. Click New deployment to create a new deployment.

  2. Select Similarity search.

  3. Select:

    • Experiment that will generate embeddings.

    • Checkpoint. The best checkpoint is the one with the lowest loss.

    • Embedding block to extract embeddings. These will be stored in the index used for similarity search.

    • Output features. These will be displayed with the similarity distance score.
      Example: An image of a lemon and the label that states it’s a lemon.

  4. Click Create.

As you do in the fruit similarity tutorial.

Enable deployment for requests

Click Enable to enable a deployed model.

Enable button

A deployed model will be accessible through the Peltarion Deployment API for forward pass queries. You can request one or several predictions at a time, within the API limitations.

The Deployment view allows you to quickly see which models are deployed and when they were deployed. A green checkmark Check mark indicates that the experiment is deployed and the date is shown in the Deployment info section.

In turn, click Disable to disable a deployed model. The deployed model will not respond with predictions while the deployment is disabled.

A deployment can be enabled and disabled several times, and can be deleted when it’s not relevant anymore. Note that you have to disable the deployment before you can delete it.

Make deployment public

To make your deployment public, toggle the Private/Public switch. This will allow you to share your results with friends and colleagues on, e.g., with the link, Twitter, LinkedIn.

Private to public


The parameter section gives a list of all the input and output features used by the deployed model.

When you submit a request to the deployed model, you have to send all the input features. The response will contain the predicted output feature for each submitted example.

The Name field refers to the name you want to use for a feature when exchanging data via the API. You can update it to something convenient to you before enabling the deployment for the first time. To change it after the deployment has been enabled once, you will need to duplicate the deployment to change it.

How to use a deployed model

Call model to request predictions

As soon as your deployment is enabled, you can start requesting predictions.

The deployment API is called by sending an HTTP POST to the URL. The request body needs to be multipart-form encoded or json.

  • The URL is the API endpoint where you submit samples.

  • The Token is required to allow the deployment to respond with predictions.
    Since the token is considered a secret, the deployment system is not meant to be shipped in the client-code (like javascript widgets, Android apps and so on).

  • Name of your deployed model’s input feature or features, as named on the Peltarion Platform.

If you are unfamiliar with REST APIs, checkout our Python package Sidekick, which makes it easy work with deployed models.

Curl example with a text input feature.
curl -X POST \
-F "text=VALUE" \
-u "9966864-xxxxxxxxxxxxxxxxxxx---3a8104f9:" \
URL and token example
Figure 1. URL and token example.

Deployment web app

For most use cases you can use the deployment web app to test your deployment in seconds.

Was this page helpful?