Skip to main content

Deploy to On-premise

This tutorial will guide you through the process when you have your own on-premise servers, and want to deploy the model cards to those servers.

Bind your devices to TensorOpera AI Platform

Bind your device to TensorOpera AI Platform.

fedml device bind $api_key

## Example output
Congratulations, your device is connected to the TensorOpera MLOps platform successfully!
Your TensorOpera Edge ID is 32314, unique device ID is 0xxxxxxxx@MacOS.Edge.Device,
master deploy ID is 31240, worker deploy ID is 31239

Check your device id from the output of the command line. Here, for example the 31240 is the master device id, 31239 is the worker device id.

You can also see your device id on TensorOpera AI Platform. onPremiseDevice.png

Deploy the model card to the on-premise device

note

Suppose you have pushed and checked the Model Card's existence on TensorOpera AI Platform Otherwise follow the previous chapter to use fedml model create and fedml model push command to create and push a local model card to TensorOpera AI Platform.

Check if the model card is uploaded to TensorOpera AI Platform by clicking the "Deploy" -> "My Models" tab on the TensorOpera AI Platform dashboard, then click the "Deploy" button on the UI. CheckModelCard.png

In the deployment page, select one master device and some worker (>=1) devices to deploy the model card. you can also configure the deployment settings, like the resource allocation, the number of replicas, etc.

selectOnpremDev.png

Click the deploy button after you select the corresponding options. After few minutes, the model will be deployed to the decentralized serverless GPU cloud. You can find the deployment details in the Deploy -> Endpoints tab in the TensorOpera AI Cloud dashboard. DeployFinished.png