fix: add embedding model as customization target

[ryan-lempka]

Problem

Running the embedding fine-tuning tutorial fails at nemo/data-flywheel/embedding-finetuning/2_finetuning_and_inference.ipynb when executing the customization job creation:

customization = client_with_wandb.customization.jobs.create(
    name="llama-3.2-1b-embed-sft",
    ...
)

Error message:

UnprocessableEntityError: Error code: 422 - {'detail': 'Model llama-3.2-nv-embedqa-1b@v2 is not enabled for customization. Contact your administrator to enable it.'}

The llama-3.2-nv-embedqa-1b-v2 embedding model is disabled by default in NeMo Microservices Helm chart v25.10+.

To Confirm Disabled Embedding Model

Pull helm chart

helm pull nmp/nemo-microservices-helm-chart --version 25.10.0 --untar

Check embedding model default configuration

grep -A 5 "nvidia/llama-3.2-nv-embedqa-1b@v2:" nemo-microservices-helm-chart/values.yaml | head -8

nvidia/llama-3.2-nv-embedqa-1b@v2:
  # -- The name for target model.
  name: llama-3.2-nv-embedqa-1b@v2
  # -- The namespace for target model.
  namespace: nvidia
  # -- Whether to enable the model.
  enabled: false

Solution

• Added a single command to the README that registers the embedding model with NeMo Customizer.
• The command uses the correct Kubernetes label selector (app.kubernetes.io/name=nemo-customizer), is idempotent and safe to run on existing deployments, and preserves all other registered models.