List of deployments

A Deployment is a control center with several features that aim at supervising a ModelVersion while it performs inferences. As explained here, a Deployment is instantiated by deploying a ModelVersion from the Private Model Registry of an Organization. It means that a given ModelVersion can have many related Deployment.

For a Computer Vision team that runs several models in production, the Deployments list is the entry point of the monitoring: it tells you at a glance which Deployment receives inferences, which one went silent, and which one is under review, before you open any of them.

1. Access the Deployments list

From the left navigation bar, click on Deployments to access the list of all the Deployment of the current Organization.

If no Deployment exists yet, the list displays No deployments. To create one, deploy a ModelVersion from the Private Model Registry, as detailed in Deploy a ModelVersion.

2. Display the list

The Deployments list can be displayed in two ways. Use the two buttons at the top left of the list to switch between them.

A. Table view

The Table view is the most complete one, as it displays one row per Deployment, with the following columns:

• Name: the name of the Deployment, a link to its overview, with its identifier below, which is useful when you work with the SDK, and an icon telling its Version Type. A green Campaign badge is displayed when a Prediction Review Campaign exists on the Deployment: clicking on it opens the Task list of this campaign (see Prediction Review Campaign).
• Model version: the ModelVersion deployed on the Deployment, a link to its overview.
• Shadow model version: the ModelVersion deployed as a Shadow Model on the Deployment, if any. It displays "-" otherwise.
• Predictions: the number of Prediction already logged in the Deployment.
• Latency: a graph displaying the latest latency trend of the inferences.
• Last inference: how long ago the last Prediction was logged, which is the quickest way to spot a Deployment that does not receive any inference anymore. It displays "-" when nothing has been logged yet.
• Created at: the creation date of the Deployment.
• Members: the avatars of the members of the Deployment.

B. Grid view

The Grid view displays one card per Deployment, which is handy to recognize your Deployment from their images. Each card shows a thumbnail of the images predicted by the Deployment, its Version Type, the number of PredictedAsset logged, its Name, and its Campaign badge if a Prediction Review Campaign exists. The card of a Deployment that has not received any inference yet has no thumbnail and displays "0 Assets".

3. Find a Deployment

When you have plenty of Deployment, three tools help you find the one you are looking for. They work the same way in the Table view and in the Grid view:

• The Search Bar at the top of the list filters the Deployment by name.
• The Order by dropdown sorts the list by Created At (default) or by Name, and the arrows button on its left switches between ascending and descending order. In the Table view, you can also click on the Name column header to sort by name.
• The Version Type buttons next to the sorting options (Classification, Object Detection, Segmentation, Mask, Point, Keypoint, and Line) only keep the Deployment of the selected Version Type. Click on All to display them all again.

The filter and the sorting are kept in the address of the page, so you can share the filtered list with a teammate by sharing the link.

4. Open a Deployment

To access a particular Deployment, click on its Name (or on its card in the Grid view): you land on its Deployment overview, detailed here.