Share via

Can't deploy a real time endpoint in azure Auto ML

Guilherme Takata 1 Reputation point
2021-08-05T17:55:51.087+00:00

I've trained a model using Auto ML and I want to try to deploy it to a Kubernetes service, I've got a pretty simple inference cluster. But no matter how low I set the CPU and memory reserve capacity it always says there isn't enough resources to deploy. Do I need to upgrade my cluster or is this a known issue with a current solution? It seems to me to be quite random as other times I was able to deploy other models with no problems at all. And even if I try to deploy it as a container instance it gets stuck in transitioning state indefinetely . Hope someone might clarify this issue and propose some kind of workaround.

Azure Machine Learning
Azure Machine Learning
An Azure machine learning service for building and deploying models.
2,728 questions

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. YutongTie-MSFT 48,586 Reputation points
    2021-08-10T16:10:32.567+00:00

    Hello,

    Thanks for the details, I just found a same issue for this abnormal computer resource issue. The solution of the customer isrewriting all the code in Python and using the AKSWebService.update_endpoint method to update the endpoint without having to delete it each time (which was happening using the Model.deploy method).

    https://learn.microsoft.com/en-us/answers/questions/249335/azure-machine-learning-update-realtime-endpoint.html

    Could you please take a look and have a try?

    Regards,
    Yutong