Deploy a model on dedicated resources using the Dashboard¶

Requirements

🔒 You need a platform account with full access level.

The Dashboard offers the possibility to deploy an AI module persistently on the platform resources.

✅ Pros	❌ Cons
Low latency in predictions, Zero configuration needed as the model is deployed internally.	You are consuming resources even when not actively making predictions.

1. Choose your module and deploy¶

The first step is to choose a module from the Dashboard. There you will be able to find all the modules ready to be used under the tag Inference.

We will first start doing a simple prediction using the popular YOLO module.

In the module page, click on the option Deploy > Inference API (dedicated). You will be show a configuration page where the option DEEPaaS is selected. You can directly click on Quick submit as you don’t need to configure anything else.

2. Making a prediction¶

In the Deployments tab, go to the Modules table and find your created deployment. This deployment offers the ability to perform predictions either via an API or an UI:

2.1 API prediction¶

Click the Quick access to access the DEEPaaS UI.

In the POST .../predict method:

click in the right button Try it out,
upload a image (eg. a bear image) to the files parameter,
click in Execute.

You will be returned a JSON with the results:

{
"status": "OK",
"predictions": [
    "[\n  {\n    \"name\": \"bear\",\n    \"class\": 21,\n    \"confidence\": 0.93346,\n    \"box\": {\n      \"x1\": 109.39322,\n      \"y1\": 26.39718,\n      \"x2\": 627.42999,\n      \"y2\": 597.74689\n    }\n  }\n]"
]
}

Notes

Sometimes the first prediction you make can be a little slow. This is because the AI model has to be download and/or loaded. Subsequent predictions should run much faster.

2.1 UI prediction¶

Under Info you will also be able to find the UI endpoint. This endpoint offers a fully fledged Gradio UI to perform inference.