APIConnectionError on Python 3.11.9, SK 1.2.0

[giobb]

Hi everyone,

I need help on this one. I have a similar C# codes that works just fine using the same settings and Managed Identity. Initially I thought MI was the culprit but switching to key does not help either.

Here is the entire block. I have applied the trick on using base_url instead of of endpoint from another thread discussing same error. Not doing so results in 404.

[moonbox3]

Hello, @giobb, thanks for reaching out.

You'll want to configure your endpoint to be: https://certain-ep.openai.azure.com/ (without the /openai/). Additionally, please use the endpoint keyword argument, instead of the base_url keyword argument. The base_url keyword argument is assuming you're passing in the base_url like: https://certain-ep.openai.azure.com/openai/deployments/gpt-4o.

As part of initializing the AzureChatCompletion service, we create the base_url from the endpoint and the chat_deployment_name such as:

https://certain-ep.openai.azure.com/openai/deployments/gpt-4o

[moonbox3]

@giobb, one debugging tip for the future is if you enable debug level logging, you'll be able to see the API call made, and would be able to then see that the URI used to make the API call to Azure OpenAI could be misconfigured.

In your script you can add:

import logging

logging.basicConfig(level=logging.DEBUG)

In the logs/console, you'd see something like:

Request: POST https://certain-ep.openai.azure.com/openai/openai/deployments/gpt-4o/chat/completions?api-version=2024-05-01-preview

Hope that helps if you run into any 404 issues later on.

[giobb]

@moonbox3 Thanks for the time and suggestion. Using endpoint (without /openai) with ad_token or api_key but still I get the same issue. Could this be firewall, other network issues?

[moonbox3]

Hi @giobb, I am using the same code as you and I am not experiencing any issues -- meaning, I am passing in my endpoint, and an ad_token. Can you confirm that you have the appropriate role on your Azure OpenAI resource? See here.

chat_service = AzureChatCompletion(
    service_id=service_id,
    endpoint="https://<resource>.openai.azure.com/",
    ad_token=ad_token,
)
kernel.add_service(chat_service)

If you do have the correct role, and you're getting a 404 still, can you confirm via debug logging that your POST request is going to the properly formatted final endpoint? Similar to what I pasted above?

Request: POST https://<resource>.openai.azure.com/openai/deployments/gpt-4o/chat/completions?api-version=2024-05-01-preview

[giobb]

Yup just double-checked. I tried even with api key. I have a similar construct in C# and it's working fine. That's the most frustrating part.

[giobb]

We'll try to upgrade to the latest SK version as you suggested. Thanks again.

[moonbox3]

For context: this was solved with an updated api version to the correct format (YYYY-MM-DD) like 2024-06-01.