How to deploy a ML model to azure which requires an entire repository to run

Question

Hi, I was following guide https://medium.com/aiguys/model-deployment-on-azure-as-a-web-app-b7bb5599bfbf
this worked, but in my actual use case I am deploying a model which is made up of 4 different ML models.

this means that my code is rather complicated and not something I can do in a simple "scoring.py" file.

I have an entire repository with a function that when called takes care of everything, my issue is how can I deploy this model as an online service which simply gets an image and then runs the code in the repository?

I uploaded this repository to https://dev.azure.com/ and also git cloned it into my ML workspace notebook.

the repository also already has the ML models to simplify things, I tried searching online but there is practically no use cases like this (which is quire surprising) so maybe I'm trying to use the wrong service.

Would really appreciate some help, thx!

Answer 1

Eitan Sharabi If you need to deploy multiple models behind an endpoint as a single deployment you can try this example from the azureml-samples repo and check if this setup works in your case. The whole end to end setup can be done with this script from the same repo.

If you are looking to deploy something behind an endpoint that actually run a code or a script in a repository, then I think the endpoints will probably timeout if your scoring script is setup to make such a call. Having the trained models in your repository can be loaded with the multi model deployment mentioned above. I would recommend trying the sample and then using the setup iteratively for your models. I hope this helps!!