in core/realtime.py [0:0]
def turn(self):
"""Finalises the pending request and creates a new one."""
self._model_done.clear()
# This is a model turn. Finish the current prompt and start a new one.
self._current_generate_output = self._pending_generate_output
self._prompt.finalize_pending()
# Prepare a new pending task for the next turn. This will do all the
# pre-processing ahead of time whenever possible.
p = debug.TTFTSingleStream('Model Generate', self._generation)
stream_content = p(self._prompt.pending())
self._pending_generate_output = processor.create_task(
context.context_cancel_coro(
self._generate_output(stream_content),
)
)