Function initialize_monitor_manager

internlm/monitor/monitor.py:213–226

(job_name: str = None, alert_address: str = None)

@contextmanager
def initialize_monitor_manager(job_name: str = None, alert_address: str = None):
    if alert_address is not None:
        try:
            monitor_manager.start_monitor(job_name=job_name, alert_address=alert_address)
            monitor_manager.handle_sigterm(alert_address=alert_address)
            send_alert_message(address=alert_address, message=f"Training in {socket.gethostname()} is starting.")
            yield
        finally:
            send_alert_message(
                address=gpc.config.alert_address, message=f"Training in {socket.gethostname()} completed."
            )
            monitor_manager.stop_monitor()
    else:
        yield