Deployments

To access and create a deployment, navigate to AI > Deployments.

Creating deployments

To create a deployment, click the blue plus button. This opens a form:

• Name: Name of the deployment.
• Model config: Model config in which the desired model was created.
• Model: Trained model.
• AI Server: Server on which to execute the model.

Click Create to continue.

Configuring deployments

To access the configuration of a deployment, click an entry in the list.

General

General configuration options for a deployment:

• Name: Name for the deployment.
• Description: Description for the deployment.
• Switch to latest model: Always use the most recently generated model for .
• Deployment host name: Zabbix host to which the reconstruction loss is sent.
• Loss item key prefix: Zabbix item key prefix to which the function name is appended. For example: if the prefix is example.loss and the selected loss function is MSE, metrics are sent to example.loss.mse.
• Model: Selected model.
• AI Server: Select an AI Server. AI Servers are configured by DataForge managers in the server configuration.
• Loss functions: Select the loss functions to use.

The host and items are configured in Zabbix and used to receive the data produced by the model. The host name and item key must be provided in the deployment configuration. The calculated loss data can then be used in Zabbix to trigger actions at defined thresholds or for other calculations.

Inference loss

This diagram shows the calculated data based on the selected loss function. This is the same data transmitted to Zabbix. The time period and loss function can be selected above the graph, which updates accordingly.