Manage and monitor deployments

Check the deployment name, description, state, selected version, primary metric, and update date before choosing an action.

Activate only when the model should accept traffic. Before deactivation, identify consumers and communicate the interruption.

Update the name or description without implying that metadata edits change model behavior. If an AI-generated description is offered, review and correct it before saving.

Hide reduces list clutter but does not stop traffic. Enable Show hidden to find and restore hidden rows.

Confirm the deployment and consumer impact before deletion. Use deactivation first when you need a reversible serving change.

Review available legacy or model versions and the associated checkpoint, such as the best or final checkpoint, when those choices are provided.

When shown, check the ML backend or serving profile associated with the version before selecting it.

Use version comparison to review metrics and model context. Do not choose a version from one headline number alone.

Make the change only after confirming compatibility and coordinating any active traffic. Run a representative test after the switch.

Use Total API Calls and Active Deployments as workspace-level orientation, then filter before drawing conclusions about one deployment.

In Traffic analysis, select the intended deployment or the aggregate view needed for the question.

Select an available period such as 7, 14, 30, 60, or 90 days, this month, or this year. Keep the same period when comparing charts.

Average Latency Per Day shows response-time movement; API Calls Per Day shows traffic volume. A change in one can affect how the other should be interpreted.

Generate the deployment usage report for the selected period, review its context, and store it according to workspace policy.