Python: Bug: invoke_stream method in chat_completion_agent.py not returning token usage data

[sdsmt05]

Describe the bug

When calling the invoke_stream method in chat_completion_agent.py the usage information isn't being sent back in the final chunk.

It appears the issue is with lines 453-460 in chat_completion_agent.py

                if (
                    role == AuthorRole.ASSISTANT
                    and response.items
                    and not any(
                        isinstance(item, (FunctionCallContent, FunctionResultContent)) for item in response.items
                    )
                ):
                    yield AgentResponseItem(message=response, thread=thread)

if response.items is empty, it skips yielding back the usage information.

To Reproduce

Call the invoke_stream method.

Expected behavior

The streaming usage information should be returned back in the final chunk.

Screenshots

Current behavior (showing zero token usage):

Desired behavior (this was achieved by commenting out line 455 in chat_completion_agent.py:

                if (
                    role == AuthorRole.ASSISTANT
                    # and response.items
                    and not any(
                        isinstance(item, (FunctionCallContent, FunctionResultContent)) for item in response.items
                    )
                ):
                    yield AgentResponseItem(message=response, thread=thread)

Platform

• Language: Python
• Source: Semantic Kernel v1.29.0
• AI model: gpt-4o-mini-2024-07-18
• IDE: VS Code
• OS: Mac

Additional context

While commenting out line 455 in chat_completion_agent.py yielded the usage information, it did also include additional "empty" partial responses.

[moonbox3]

Thanks for filing, @sdsmt05. We'll get this handled ASAP.