Make Your Container Read the Model
Overview
Deploying an edition (see Upload and Deploy) delivers the verified model file to the device and restarts your inference container — but the apply pipeline only restarts the container so it re-reads the file. It does not create or mount a volume for you.
Your inference container must already be configured to mount the device's secured volume before the first deploy. This page covers that one-time container setup.
WEDA automatically creates and manages the secured volume for you as part of deployment — there is no API to provision or configure it directly. Your container connects to it as a read-only volume using the connection details your WEDA deployment provides.
The secured volume's internal management is not exposed as a public REST API in v1.1.0. Mounting it in your container's own deployment definition, as shown below, is the supported approach today.
Where the Model File Lands
Once a deployment reaches DEPLOYED, the verified file is available inside the secured volume.
Use the edgeFilePath value returned in the deployment result (see
Check Deployment Result) as the path within
the mounted volume — it is already relative, in the form
/ai-models/<modelName>/<edition>/<fileName>. Don't construct or assume any other path; the
underlying mount structure is managed entirely by WEDA and may change between releases.
Mount the Volume in Your Container
Configure your container's deployment definition (Docker Compose service) to mount the secured
volume as a read-only NFS volume, then load the model from <mountPath><edgeFilePath> in your
inference code:
# docker-compose.yml excerpt
services:
infer-svc:
volumes:
- type: volume
source: secured-volume
target: /models
volumes:
secured-volume:
driver: local
driver_opts:
type: nfs
o: "addr=<secured-volume-ip>,vers=3,nolock,ro"
device: "<secured-volume-export-path>"
<secured-volume-ip> and <secured-volume-export-path> are fixed for your deployment — contact
your WEDA administrator or Advantech support for the values to use, and reference them directly in
your Compose file rather than discovering them dynamically. They do not change once provided.
Deploy this Compose file as a WEDA Container Stack so the mount configuration is managed through WEDA rather than run manually on the device.
Verify
Confirm the file is visible inside your container at its configured mount path:
docker exec <infer-svc> ls -la /models<edgeFilePath>
If the file is missing, the secured-volume mount is blocking every AI model download for that device — resolve the mount before retrying the deployment.
Related
- Deploy an edition and check its rollout status: Upload and Deploy
- Package your inference workload as a container: Container Stack