Search code examples
azure-cognitive-servicesazure-openai

Can't deploy Azure Open AI models due "No quota is available for this deployment. You can request for more quota."

I wanted to test the gpt-35-turbo-16k model, so I attempted to deploy it. However, I received a message stating that there was no quota available. I initially assumed that it might be due to an existing deployment of the gpt-35-turbo model, so I deleted it. Unfortunately, I encountered the same issue even after removing the previous deployment. Furthermore, I tried redeploying the gpt-35-turbo model but was unable to do so due to the same problem. enter image description here

This is my actual quota: enter image description here

And those are the models that I have deployed: enter image description here

Edit:

I have created two new instances after encountering the issue. The first one was just to see if the error disappears with a fresh instance, and the second one was to try a different region, but unfortunately, neither of them worked. I also purged instances in case there was another one that I didn't remember, but that wasn't the case. I am still facing the same issue. Another weird thing that occurred today was that the gpt-35-turbo-16k model disappeared from the list of deployable models.

Solution

  • I opened a support ticket with Microsoft and they gave me a workaround.

    In the meanwhile, as a mitigation, would you please provide read(microsoft.cognitiveservices/locations/usages/read) permission in a subscription level. Usages API request on Open AI Studio pulls current usage of OpenAI from the subscription level.

    Seems like the problem is related to permissions, if you deploy the model with an account with subscription level permissions you will be able to deploy the models. In my case my user has resource group level permissions so I think is unable to "read" if there is enough quota available to do the deploy.

    I think this behavior is no correct but unfortunately seems like at least this Microsoft representant isn't agree with me.

    I had a chance to further discuss this issue with the product team. I would like to clarify that the entire quota stack is subscription level based for Open AI. So being an owner/ contributor for the resource/ resource group wouldn't make a difference in this case. You need to have the permission in subscription level to make quota changes. Either contributor or a Reader works for subscription level role. Product team are working on showing a better error message rather than "No quota..." and also better documentation.

    I hope Microsoft realize this mistake and we will be able again to deploy models without the need of subscription level permissions.

    Edit 07/03/2024: Now Microsoft show some extra information about quota management in its documentation (someone asked so added to my answer).

    enter image description here https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/quota?tabs=rest#prerequisites