@yuichiromukaiyama note that our GitHub actions markdown-link-check step is currently failing -- it isn't related to your changes. We're trying to investigate as it is blocking our pipelines.

Thank you so much！

@yuichiromukaiyama your change went in just in time for the latest release. Thank you for your contribution!

I am still getting error..I am using "0.5.1.dev0", I see latest code ,but not sure issues is resolved completely

def _get_metadata_from_chat_choice(self, choice: Union[Choice, ChunkChoice]) -> Dict[str, Any]:
"""Get metadata from a chat choice."""
return {
"logprobs": getattr(choice, "logprobs", None),
}

@veeruhai
Do you fall under this problem? I think there are cases where logprobs do not exist for APIs other than ChatCompletion.
The problem in the following case is that even though APIKEY uses ChatCompletion, the processing flow is TextCompletion. I thought I would raise this as a separate issue. However, since there are multiple countermeasures, I wrote it in the discussion.

↓

#5011

@yuichiromukaiyama ,
for me none of python notebooks working, I am using AzureOpenAI service, followed your workaround and changed PromptTemplate to ChatPromptTemplate in kernel .now I am able to run 00-getting-started notebook and basic planner notebook. Still facing same error for other planner type.

@veeruhai

I investigated the issue and I believe it is the same as the one I described in the "Discussion".

Both use PromptTemplate, and the error seems to be occurring in a class that I have not modified.

The problem can be solved by changing the PromptTemplate used in this class to ChatPromptTemplate.

https://github.com/yuichiromukaiyama/semantic-kernel/blob/9cedc95b4ce9bf7aff6ae16892df757076ca1bbb/python/semantic_kernel/planning/sequential_planner/sequential_planner.py#L62

I believe that the sequential planner also performs the following process:

https://github.com/yuichiromukaiyama/semantic-kernel/blob/9cedc95b4ce9bf7aff6ae16892df757076ca1bbb/python/semantic_kernel/kernel.py#L171-L172

To solve the root cause, I plan to issue a PR to make the following changes:

↓

In the following section, kernel.register_semantic_function is called in _init_flow_function of SequentialPlanner.

https://github.com/yuichiromukaiyama/semantic-kernel/blob/9cedc95b4ce9bf7aff6ae16892df757076ca1bbb/python/semantic_kernel/planning/sequential_planner/sequential_planner.py#L74-L75

Since the use of PromptTemplate is enforced in Planner, TextCompletionClientBase is used in the following section:

https://github.com/yuichiromukaiyama/semantic-kernel/blob/9cedc95b4ce9bf7aff6ae16892df757076ca1bbb/python/semantic_kernel/kernel.py#L754

There is TextCompletionClientBase, which is inherited by OpenAITextCompletionBase.

https://github.com/yuichiromukaiyama/semantic-kernel/blob/9cedc95b4ce9bf7aff6ae16892df757076ca1bbb/python/semantic_kernel/connectors/ai/open_ai/services/open_ai_text_completion_base.py#L29-L30

OpenAITextCompletionBase also has the corresponding logprobs code.

https://github.com/yuichiromukaiyama/semantic-kernel/blob/9cedc95b4ce9bf7aff6ae16892df757076ca1bbb/python/semantic_kernel/connectors/ai/open_ai/services/open_ai_text_completion_base.py#L144

This will be rewritten as optional in the same way.

@veeruhai I have issued a pull request.
#5071