diff --git a/src/mnemosyne/gateway.py b/src/mnemosyne/gateway.py index a7d08b4..a44893a 100644 --- a/src/mnemosyne/gateway.py +++ b/src/mnemosyne/gateway.py @@ -1881,7 +1881,12 @@ def create_app( exc_info=True, ) - def _display_turn_status(usage: dict, session: Session) -> None: + def _display_turn_status( + usage: dict, + session: Session, + incoming_bytes: int = 0, + outgoing_bytes: int = 0, + ) -> None: """Post-response status line with cache hit rate.""" sid = session.id ts = session.token_state @@ -1898,8 +1903,8 @@ def create_app( effective_tokens=effective, cache_read=cache_read, cache_create=cache_create, - incoming_bytes=0, # filled by telemetry record_request - outgoing_bytes=0, + incoming_bytes=incoming_bytes, + outgoing_bytes=outgoing_bytes, ) cap_str = "" @@ -2078,7 +2083,12 @@ def create_app( # Post-response: track usage, display status if usage: _check_token_cap(usage, session) - _display_turn_status(usage, session) + _display_turn_status( + usage, + session, + incoming_bytes=incoming_bytes, + outgoing_bytes=outgoing_bytes, + ) _update_fidelity_pressure(usage, session) emit_event( "response_observed", @@ -2134,7 +2144,12 @@ def create_app( if usage: _check_token_cap(usage, session) - _display_turn_status(usage, session) + _display_turn_status( + usage, + session, + incoming_bytes=incoming_bytes, + outgoing_bytes=outgoing_bytes, + ) _update_fidelity_pressure(usage, session) emit_event(