gitv2

Commit: 73f0333

Date: July 11, 2025 at 1:07 PM

For example, for a typical function tool call sequence, we need

1) runOneStep (= 1 LLM call)

- client calls LLM with info about available tools.

- model responses with the picked function tool and args.

- client calls the function with the args and yield the event.

2) next iteration of the for loop.

3) runOneStep (= 1 LLM call)

- client calls LLM with the function call result (along with previous history)

- model responses based on the history and the function call result.

4) next iteration or get out of the loop.

• fix misuse of yield in loop.

(When looping, we need to check yield's return value and get out of the loop)

• if partial content arrives, now, yield an error and exit.

(note: in python it threw an error, but it is recently changed so it just logs and exit)