Intelligent Virtual Assistant
This section will walk through an end-to-end workflow deployment using the example software stack components previously described.
Ensure that the two nodes provisioned from the previous Hardware Requirements section are accessible.
One of the VMIs will be used for the training pipeline.
The second Kubernetes cluster node will be used for the inference pipeline.
SSH into the training VMI (this is the VMI without Cloud Native Stack or Kubernetes installed).
-
ngc registry resource download-version "nvaie/intelligent-virtual-assistant-training:0.1"
Switch to the training directory.
cd intelligent-virtual-assistant-training_v0.1/
Make the set-up script executable.
chmod +x ./run.sh
Run the setup script.
sudo ./run.sh <YOUR-API-KEY>
NoteThe installer may fail if dpkg does not run cleanly or entirely during the instance provisioning. If this occurs, run the following command to resolve the issue, then retry the installation.
sudo dpkg --configure -a
From a browser, navigate to the Jupyter Notebook URL displayed once the setup script is completed. It is part of the CUSTOM STATES output, e.g.
RUN: { "services": [ { "name": "notebook", "url":"**http://<External-IP>/notebook**" } ]}
Select and run through the Jupyter Notebooks, starting with the Welcome Notebook.
The Training Deployment steps are complete
As a part of the workflow, we will be demonstrating how to deploy the packaged workflow components as a Helm chart on the previously described Kubernetes-based platform. We will also demonstrate how to interact with the workflow, how each of the components in the pipeline work, and how they all function together.
This includes an example of how to securely send requests to the inference pipeline, using Envoy set up as a proxy to authenticate and authorize requests sent to Triton, and Keycloak as the OIDC identity provider. For more information about the authentication portion of the workflow, refer to the Authentication section in the Appendix.
First, configure Keycloak according to the instructions provided in the Appendix.
Note down these six fields for the Deployment workflow.
Client-id
Client-secret
Realm Name
Username
Password
Token_endpoint
SSH into the inference/deployment VMI.
Once Keycloak has been configured, run the following command on your system via the SSH console to get the access token (replace the
TOKEN_ENDPOINT
,CLIENT_ID
,CLIENT_SECRET
,USERNAME
andPASSWORD
fields with the values previously created).curl -k -L -X POST '<TOKEN_ENDPOINT>' -H 'Content-Type: application/x-www-form-urlencoded' --data-urlencode 'client_id=<CLIENT_ID>' --data-urlencode 'grant_type=password' --data-urlencode 'client_secret=<CLIENT_SECRET>' --data-urlencode 'scope=openid' --data-urlencode 'username=<USERNAME>' --data-urlencode 'password=<PASSWORD>' | json_pp
For example:
curl -k -L -X POST 'https://auth.your-cluster.your-domain.com/realms/ai-workflows/protocol/openid-connect/token' -H 'Content-Type: application/x-www-form-urlencoded' --data-urlencode 'client_id=merlin-workflow-client' --data-urlencode 'grant_type=password' --data-urlencode 'client_secret=vihhgVP76TgA4qDL3c5jUFAN1gixWYT8' --data-urlencode 'scope=openid' --data-urlencode 'username=nvidia' --data-urlencode 'password=hello123' | json_pp
This will output a JSON string like below
{"access_token":"eyJhbGc...","expires_in":54000,"refresh_expires_in":108000,"refresh_token":"eyJhbGci...","not-before-policy":0,"session_state":"e7e23016-2307-4290-af45-2c79ee79d0a1","scope":"openid email profile"}
Note down the
access_token
, this field will be required later on in the workflow, within the Jupyter notebook.Now we’re ready to deploy the intelligent virtual assistant application.
Ensure the NGC CLI is configured.
ngc config set
-
ngc registry resource download-version "nvaie/intelligent-virtual-assistant-deployment:0.1"
Switch to the transcription Helm chart directory.
cd intelligent-virtual-assistant-deployment_v0.1/helm_charts
Install Haystack Helm chart.
helm -n cciva install haystack haystack --create-namespace
Note down the haystack url from the output of the Helm install.
Install the Rasa Helm chart.
helm -n cciva install rasa rasa/ --set ngcCredentials.password=<NGC_KEY> --set haystackUrl=<HAYSTACK_URL>
Note down the Rasa URL from the Helm chart output.
ImportantIf you have already installed the Audio Transcription Riva Helm Chart, ensure you delete it before proceeding if you only have a single GPU attached to your VMI.
cd ~/audio-transcription-deployment_v0.1/helm_charts helm del transcription -n riva
Install Riva via Helm.
helm -n cciva install riva riva/ --set ngcCredentials.password=<NGC_KEY> --set workflow.keycloak.keycloakrealm=<WORKFLOW_REALM> --create-namespace --set haystackUrl=<HAYSTACK_URL> --set rasaUrl=<RASA_URL>
Reference the output from the Helm install and launch the Jupyter Notebook from a browser once the transcription pods are running.
Validate that all of the iva pods are running
kubectl get pods -n cciva
NoteIt can take about half an hour for the Riva pod to initialize
Once all pods are running, access the Jupyter Notebook for instructions to run the workflow