Overview
Use AI Model Management to upload model files, manage their versions and deploy a selected version to WEDA Nodes. Keep the model file separate from the container image that provides your inference application's runtime.
In WEDA Console, use the Org's AI Models page to manage Models and Editions, upload files and select deployment targets. Use a device's AI Models tab to inspect its model deployments. The APIs support integrating these operations into your own application or deployment pipeline.
Model, Edition and Deployment
The feature uses three related resources:
| Resource | What it represents | Example |
|---|---|---|
| Model | A named group of model files and their versions within an Org. Creating a Model does not upload a file. | defect-detector |
| Edition | One version of a Model's file. Upload and verification must complete before it is ready for deployment. | Edition 1-0-3 of defect-detector |
| Deployment | Delivery of a selected Edition to target devices, with a result tracked for each device. | Deploy defect-detector / 1-0-3 to the WEDA Node on an inspection station. |
After an Edition's file is uploaded and verified, its contents cannot be replaced. Register another Edition for a different file; descriptive metadata can still be updated.
For example, keep successive versions of your inspection model under defect-detector. Select the Edition you intend to deliver, then check the result on each target device. Your inference application must be configured to read that file and load it with a compatible runtime.
The Full Picture
The workflow has two delivery phases:
- Upload: Create the Model and Edition, transfer the file using resumable upload and wait for server-side SHA-256 verification. A completed transfer alone does not mean the Edition is ready to deploy.
- Deploy: Select an uploaded Edition and target devices, configure the deployment options and check the device-reported results. Submitting a request alone does not confirm that a device has finished.
Your inference container reads the delivered file through its configured volume mount. Prepare that container configuration before expecting the application to use the model.
Delivering a File and Using a Model
Check the delivery and the application separately:
- Device deployment: Did the selected Edition reach the target device and complete the configured deployment actions? Check the deployment state and any reported failure details.
- Application use: Can the container read the file, and did the inference application load the intended Edition? Check the application's own behavior or output.
Deployment options can include replacing the model file and restarting a specified container. The container and its volume mount must already be configured; a restart by itself does not prepare the mount or confirm correct inference.
A successful model deployment does not by itself confirm that your running application has loaded the new model. Verify both the device's deployment result and the application's use of the intended Edition.
Next Steps
| Guide | Use it to |
|---|---|
| Upload an AI Model | Prepare the file, create a Model and Edition, upload and verify it, and manage the Cloud records. |
| Deploy an AI Model | Choose an Edition and devices, configure deployment actions, monitor results, retry or remove device deployments. |
| Make Your Container Read the Model | Configure the volume mount and verify that your application can read and use the deployed model file. |
Use Container Stack Management to manage the containerized application that runs your inference workload.