FEAT: SelectorGroupChat could using stream inner select_prompt

[SongChiYoung]

Why are these changes needed?

This PR updates SelectorGroupChat to support streaming mode for select_speaker.
It introduces a streaming argument — when set to True, select_speaker will use create_streaming() instead of create().

Additional context

Some models (e.g., QwQ) only work properly in streaming mode.
To support them, the prompt selection step in SelectorGroupChat must also run with streaming=True.

Related issue number

Closes #6145

Checks

• I've included any doc changes needed for https://microsoft.github.io/autogen/. See https://github.com/microsoft/autogen/blob/main/CONTRIBUTING.md to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[ekzhu]

Can we address this first? #6161. Otherwise the streaming option won't actually stream to run_stream.

[ekzhu]

_message actualy has the type str or CreateResult.

[SongChiYoung]

You’re right, that was my mistake.
I misunderstood how create_stream() works and thought the log message was being leaked.

But actually, the last message is always a CreateResult, and the if block just skips over LLMStreamEndEvent, so it worked by accident.
Thanks for pointing it out!

[SongChiYoung]

@ekzhu

Can we address this first? #6161. Otherwise the streaming option won't actually stream to run_stream.

I’ll take a look at #6161.
That said, in my case, I confirmed that stream was working by adding some print statements inside the create_stream() function at Openai Client

And, just change

            response = await self._model_client.create(messages=select_speaker_messages)

to

            if self._streaming:
                message: CreateResult | str = ""
                async for _message in self._model_client.create_stream(messages=select_speaker_messages):
                    message = _message
                if isinstance(message, CreateResult):
                    response = message
                else:
                    raise ValueError("Model failed to select a speaker.")
            else:
                response = await self._model_client.create(messages=select_speaker_messages)

So I don't think there would be any problem solving that issue —
-> well, except maybe some merge conflict later.

BTW, what kind of help do you expect for that issue?
There’s no code or draft linked yet.

Do you want me to take the lead on it?

[ekzhu]

Yes, please address #6161 and then this one.

[SongChiYoung]

@ekzhu I resolved merge conflict.

[codecov]

Codecov Report

Attention: Patch coverage is 53.84615% with 6 lines in your changes missing coverage. Please review.

Project coverage is 77.52%. Comparing base (bb792b0) to head (ce0ac44).

Files with missing lines	Patch %	Lines
...gentchat/teams/_group_chat/_selector_group_chat.py	53.84%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6286      +/-   ##
==========================================
- Coverage   77.55%   77.52%   -0.03%     
==========================================
  Files         202      202              
  Lines       14759    14770      +11     
==========================================
+ Hits        11446    11451       +5     
- Misses       3313     3319       +6     

Flag	Coverage Δ
unittests	77.52% <53.84%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[ekzhu]

See how emitting event is done in BaseGroupChatManager:

await self._output_message_queue.put(ModelClientStreamingChunkEvent(content=_message)) to emit the chunk.

[SongChiYoung]

Howabout, call _log_speaker_selection after makes total response?

Now I know about why do not use it.

like that case (adding [ and ] for more looks easy. look likely BaseGroupChatManager output) looks weird

---------- user ----------
say hello world
---------- SelectorGroupChatManager ----------
[hello]
---------- hello ----------
Hello, World!
---------- SelectorGroupChatManager ----------
Model failed to select a speaker after 3, using the previous speaker.
[hello][hello][,][ world][!][hello][,][ world][!]
---------- hello ----------
Hello, World!

Real three output is that
[hello] [hello, world!] [hello, world!]

So, howabout call await self._output_message_queue.put(ModelClientStreamingChunkEvent(content=_message)) after, set response?

[SongChiYoung]

And... without that code,
output is like it

---------- user ----------
say hello world
---------- SelectorGroupChatManager ----------
['hello']
---------- hello ----------
Hello, World!
Model failed to select a speaker after 3, using the previous speaker.
---------- SelectorGroupChatManager ----------
['hello']
---------- hello ----------
Hello, World!

so... with that line, default _log_speaker_selection is looks not works.
(however, actually that function is called, just do not shown)

Clearly I need to your help for understand that funciton.

[ekzhu]

_log_speaker_selection shouldn't be mixed in here. Keep it separate in the base class.

In this PR we are addressing the SelectorGroupChat-specific issue.

After every sequence of ModelClientStreamingEvent, it should be followed with a separate message which has the full content of the streamed token. This can be a SelectorEvent, for example.

[SongChiYoung]

------ Delete that comment ------
Cause, now understand of your direction.

[ekzhu]

@SongChiYoung generally, I think it is more important to emit the inner SelectorEvent activities rather than just streaming the acitivies.

[SongChiYoung]

@ekzhu
Thanks for all of your help.
I think, all of issue is resolved and changed from your direction.
and, I was add test case for this change.

Please check this PR.