Web App to deploy Models

1

Hi All!

I am planing to create a Web App to deploy all my Deep learning models, execute them in the cloud with any image/text (depending the problem) the user would like to input and output results.

What options do I have?

Is this a Kubernetes problem?

Thanks for your help!

Facundo

2

You could consider Seldon + Kubeflow, which has direct integration for PyTorch serving. Alternatively you could export your models to ONNX and use the MXNet model server directly, or you could then import them into TensorFlow and use TensorFlow Serving, or if they are light enough you could also consider client-side inference through ONNX.js.
Or if your use case is more lightweight and these options seem somewhat overkill you could always make a simple server with flask or any other serving framework and call your model directly, or deploy it serverless through a lambda/cloud function.

3 Likes