vllm_omni.engine.duplex.session.playback_ledger ¶
Applying one playback.ack to a duplex session.
The client reports how much of a response it has actually played; the session uses that to commit or truncate the assistant item in conversation history, and to release the response's playback state once it is fully played.
Extracted from DuplexSessionRunner because the only runner state it touched was session. It returns the events it produces rather than emitting them, so the rules can be exercised without a runner.
apply_playback_ack ¶
apply_playback_ack(
session: DuplexEngineSession, event: dict[str, object]
) -> list[DuplexEvent]
Apply one playback.ack and return the events it produces.