Create a custom runtime image for Watson Machine Learning Accelerator

Use your own custom runtime image to run workloads.

Create a custom Dockerfile using the built-in image as a base:

For example:

From cp.icr.io/cp/cpd/wml-accelerator-runtime:v4.0.0-rt23.1-py3.10
RUN <your-customization-steps>

Build and push the image:

podman build -f Dockerfile -t <your-registry>/<your-image> .

Push the image to your own registry:

podman push <your-registry>/<your-image>

For your OpenShift cluster to pull the image, update the global pull secret. For details, see: Updating the global cluster pull secret.

Update the IBM Watson® Machine Learning Accelerator CR to use the custom image:

Edit the wmla-install-info-cm config map:

 $ oc describe cm wmla-install-info-cm -n wmla-03
 ...
 Data
 ====
 ...
 my_runtime_image: cp.icr.io/cp/cpd/wml-accelerator-runtime@sha256:9c3890214eeb00e2e80e6079576cc99ede1102729db1bd7a4a9c266434e60658

Edit the Watson Machine Learning Accelerator CR as follows:

apiVersion: spectrumcomputing.ibm.com/v1
 kind: Wmla
 spec:
    version: 4.0.0
    ...
    selectedRuntimeImage:
       targetVersion: 4.0.0 
       imageRef: my_runtime_image

Wait for the update to be applied. Verify that your runtime image is set. Restart the dlpd pod if needed.