Question about Tool Call Streaming

[Lightblues]

I found that OpenAIChatCompletionsModel can only stream out text, but cannot stream out tool_call deltas. But I think streaming is an important feature for application that need instant responses.

There is a comment in the code that "Because we don't know the name of the function until the end of the stream, we'll save everything and yield events at the end", but I don't understand it. Could you clarify the reason why we cannot just yield ResponseFunctionCallArgumentsDeltaEvent (openai.types.responses.response_function_call_arguments_delta_event) in ChatCompletions API?

[rm-openai]

We get deltas that contain:

• deltas of the function name
• deltas of the function args

So e.g. if your function is called "get_weather", the first delta might have get_ and the second might have weather. So until we receive the last delta, we don't know the name of the function.

[rm-openai]

In practice this doesn't seem to be a real issue so maybe we should fix this

[Lightblues]

Thank you for your attention to this issue. I appreciate your insights and look forward to seeing progress on this issue.

[ixjx]

+1 for this feature! I also need the ability to stream tool_call deltas. This would be crucial for applications that require real-time tool invocation and instant responses.

[cm2435]

just FYI for people in this thread: ResponseOutputItemAddedEvent is the first outputted token of a streamed function call and contains the entire function name.

See https://platform.openai.com/docs/guides/function-calling?api-mode=responses#streaming