When you promote a model version to production, downstream systems often need to react automatically—updating a running service, alerting your team, or calling a webhook. Model version approval status lets you govern which versions are production-ready and trigger those downstream actions the moment a version's status changes.
Prerequisites
Before you begin, ensure that you have:
EventBridge permissions granted to Platform for AI (PAI) in your workspace. For details, see Create a notification rule
How it works
Each model version has one of three approval statuses:
| Status | Description |
|---|---|
| Pending | Default status when a version is created. The version can be deployed manually. |
| Approved | The version is approved for production use. |
| Rejected | The version is rejected from production use. |
Approved and Rejected are terminal states. Once a version reaches either status, the status cannot be changed.
The following status transitions trigger configured events automatically:
| Transition | Event triggered | Consequence |
|---|---|---|
| Pending → Approved | Yes | Configured notifications and EAS service updates fire automatically |
| Pending → Rejected | Yes | Configured notifications fire automatically |
Events only fire when a notification rule is configured for the model. For details, see Create a notification rule.
Configure event rules for model version approval
To set up event-driven automation for model version approval changes, create a notification rule and select Model Management as the Event Source.
Supported model scope
| Scope | Description |
|---|---|
| Specified models | Events apply only to the models you select |
| Models that you created | Events apply to all models you own in the workspace |
| All models in the current workspace | Events apply to every model in the workspace |
Supported event types
| Event type | Trigger condition | Notes |
|---|---|---|
| DingTalk messages | Any configured status change | Sends a real-time notification via a DingTalk chatbot |
| HTTP/HTTPS events | Any configured status change | Calls the specified endpoint; the endpoint must parse the payload based on the specification template |
| EAS service update | Status changes to Approved only | Automatically updates the specified running Elastic Algorithm Service (EAS) service |
Set the approval status of a model version
When registering a model or adding a version:
Set the Version Approval Status field to Pending or Approved during registration. For details, see Register and manage models.
For already-registered model versions:
Update the approval status directly from the model version list.
Use the Model Management OpenAPI or SDK to Configure
For more information, see Update a model version.