nemo-data-flywheel-tutorials

Tutorials for NeMo Microservices (MS) Data Flywheel, including examples for using the NeMo MS Data Store, Entity Store, Customizer, Evaluator, Guardrails, and NVIDIA NIMs.

1. Steps to run the notebook

Update Notebook Config Update the test/e2e/jupyter-notebook/config.py file with the following values:

NDS_URL = "http://nemodatastore-sample.nemo.svc.cluster.local:8000" # Data Store
ENTITY_STORE_URL = "http://nemoentitystore-sample.nemo.svc.cluster.local:8000" # Entity Store
CUSTOMIZER_URL = "http://nemocustomizer-sample.nemo.svc.cluster.local:8000" # Customizer
EVALUATOR_URL = "http://nemoevaluator-sample.nemo.svc.cluster.local:8000" # Evaluator
GUARDRAILS_URL = "http://nemoguardrails-sample.nemo.svc.cluster.local:8000" # Guardrails
NIM_URL = "http://meta-llama3-1b-instruct.nemo.svc.cluster.local:8000" # NIM
HF_TOKEN = "<your-huggingface-token>"
BASE_MODEL = "meta/llama-3.2-1b-instruct"

NOTE: if you have are already gone through steps in the QuickStart guide, you can skip steps 2-4

Install the NeMo Dependencies Ansible playbook that deploys the Jupyter server with all required NeMo dependencies enabled in values.yaml.

install:
  customizer: yes
  datastore: yes
  entity_store: yes
  evaluator: yes
  jupyter: yes

Deploy the NeMo Training Operator

kubectl create ns nemo-operator
kubectl create secret -n nemo-operator docker-registry ngc-secret \
  --docker-server=nvcr.io \
  --docker-username='$oauthtoken' \
  --docker-password=<ngc-api-key>

helm fetch https://helm.ngc.nvidia.com/nvidia/nemo-microservices/charts/nemo-operator-25.4.0.tgz --username='$oauthtoken' --password=<YOUR NGC API KEY>
helm install nemo-operator-25.4.0.tgz -n nemo-operator --set imagePullSecrets[0].name=ngc-secret --set controllerManager.manager.scheduler=volcano

Create Custom Resources (CRs) for all NeMo and NIM samples from config/samples/nemo/latest folder.

kubectl apply -f config/samples/nemo/latest

Access your jupyter server

Once the Ansible playbook has completed, the Jupyter server will be running in your cluster.

To access the Jupyter notebook, use kubectl port-forward from your local machine and launch using http://localhost:8888. The token is "token"

kubectl port-forward svc/jupyter-service -n nemo 8888:8888

Access your notebook

Once you launch the Jupyter server, the e2e test notebook is under the work directory.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!