Issue (or wrong usage) with OpenAiChatCompletionBase

[Icewater1337]

Hey. I have a kernel with an AzureChatCompletionService using the Python library.

now what I noticed is this, that when the OpenAI api returns an error, the ChatCompletion response has choices = None, and the error is in the 'model_extra' dict.

semantic-kernel/python/semantic_kernel/connectors/ai/open_ai/services/open_ai_chat_completion_base.py

Line 88 in 1322314

response = await self._send_request(settings)

However on

semantic-kernel/python/semantic_kernel/connectors/ai/open_ai/services/open_ai_chat_completion_base.py

Line 91 in 1322314

return [self._create_chat_message_content(response, choice, response_metadata) for choice in response.choices]

we can see, that it tries to iterate over response.choices nevertheless, which always gives a cant iterate over None exception and the execution fails, without ever returning the API error.

Is this intended? am I using it wrong somehow?

My package versions:

semantic-kernel==1.24.1
openai==1.68.0

Model that I call:
"gpt-4o-2024-05-13"
and api version
"2024-10-21"

I am hesitant to open an Issue yet, because I am not sure if i miss something.

[moonbox3]

Hi @Icewater1337, this looks like a bug -- I created an issue #11208 to track this fix (track along there for further progress). Thanks for bringing this up.

[Icewater1337]

Superb, thanks for the feedback and for creating the bug!