NVIDIA Fleet Command¶
NVIDIA Fleet Command is a hybrid-cloud platform for securely and remotely deploying, managing, and scaling AI across dozens or up to thousands of servers or edge devices. It is currently released in early access, see the link to apply. Jarvis can be deployed into a Fleet Command environment using the Jarvis Helm Chart located in NGC.
Application Setup¶
Setup the application within Fleet Command by selecting APPLICATIONS from the Fleet Command UI.
Select Add Application from the top left menu.
Fill out the modal popup:
Display Name:
Jarvis Speech
Description:
Jarvis Fleet Command deployment
Helm Name:
jarvis-api
Helm version: 1.1.0-beta
Helm URL: https://catalog.ngc.nvidia.com/orgs/nvidia/teams/riva/helm-charts/riva-api
Select Add.
Deployment Setup¶
Create the deployment within Fleet Command by selecting Deployments > Create Deployment from the UI.
Fill out the Name, Display Name, and select the location to deploy.
Select from the application drop-down box the name from**Display Name** from Application Setup. For example, “Jarvis Speech”.
Create secrets. At any bash shell, create the model deployment key. Fleet Command should already be configured with
modelpullsecret
andimagepullsecret
.For the model key, run:
echo -n tlt_encode | base64 -w0 dGx0X2VuY29kZQ==
Copy and paste the following into the Application Configuration dialog box (execute the shell snippet next to the password to create your encoded password). This will deploy all three services with the default models. The
nodeport
value needs to be a value betwen 30000-32767, and will be the port on which all the services are exposed.modelRepoGenerator: modelDeployKey: dGx0X2VuY29kZQ== modelDeployVolume: hostPath: path: /opt/data/jarvis artifactDeployVolume: hostPath: path: /opt/data/jarvis service: type: NodePort nodeport: 32222
To install just one type of service, add the following and adjust as needed.
jarvis: speechServices: asr: true tts: false nlp: false
Check the box I understand…
Click the Deploy button in the top right of the page. Based on the models and services to be deployed, the installation can take more than 60 minutes.
If the deployment does not show as green, consult the Fleet Command documentation for how to pull logs and perform troubleshooting of helm deploys.