Custom Container Integration

This tutorial will guide you through the process of creating a local model card using a customized Docker image from DockerHub and deploying the model card to a serverless GPU cloud.

note

This tutorial assume you have an App containerized in a Docker image, and you want to create a model card for it. If you haven't has that App containerized, you can follow the Create a Model Card tutorial to create a model card from a model config file.

Prerequisites

Install fedml, the serving library provided by TensorOpera AI, on your machine.

pip install fedml

Create a model card using a configuration file

Suppose you have a docker image on DockerHub that is ready for serving. You can create a model card by providing a configuration file. The configuration file should contain the following fields:

# Example config.yaml
inference_image_name: "nvcr.io/nvidia/tritonserver:24.05-py3" # str, Docker image name
container_run_command: "tritonserver --model-repository=/repo_inside_container" # str or list, similar to CMD in the dockerfile, or docker run command.
port: 8000                          # int, service port, currently you can only indicate one arbitrary port.
expose_subdomains: true             # bool, set to true route all the subdomains, set to true. e.g. localhost:2345/{all-subdomain}.
workspace: "./"                     # str, pacakge other files in the workspace to the model card. e.g. README.md
readiness_probe:                    # dict, the readiness probe configuration.
  httpGet:
    path: "/v2/health/ready"

Use fedml model create command to create a model card on your local machine.

fedml model create -n $my_model_name -cf config.yaml

Deploy the model to a Serverless GPU Cloud

tip

Before you start, you will need to create an account on TensorOpera AI.

Use fedml model push to push the model card to TensorOpera AI Cloud. Replace $api_key with your own API key. The API Key can be found from the profile page.

fedml model push -n hf_model -k $api_key

After you push the model card to TensorOpera AI Cloud, you can deploy the model by going to the Deploy -> My Models tab on the TensorOpera AI Platform dashboard. Click the Deploy button to deploy the model.

For this quick start tutorial, we can select the Serverless RTX-4090 option and click the Deploy button.

After few minutes, the model will be deployed to the serverless GPU cloud. You can find the deployment details in the Deploy -> Endpoints tab in the TensorOpera AI Cloud dashboard.

You may interact with the deployed model by using the curl, Python, or JavaScript command under the API tab.