Method start_heter_trainer

python/paddle/base/executor.py:2967–3009

(
        self,
        program: Program | None = None,
        scope: core._Scope | None = None,
        debug: bool = False,
        fetch_list: list[Tensor] | None = None,
        fetch_info: list[str] | None = None,
        print_period: int = 100,
        fetch_handler: FetchHandler | None = None,
    )

        )

    def start_heter_trainer(
        self,
        program: Program | None = None,
        scope: core._Scope | None = None,
        debug: bool = False,
        fetch_list: list[Tensor] | None = None,
        fetch_info: list[str] | None = None,
        print_period: int = 100,
        fetch_handler: FetchHandler | None = None,
    ) -> core.TrainerBase:
        scope, trainer = self._prepare_trainer(
            program=program,
            dataset=None,
            scope=scope,
            thread=1,
            debug=debug,
            fetch_list=fetch_list,
            fetch_info=fetch_info,
            print_period=print_period,
        )

        trainer._set_infer(False)
        trainer._gen_trainer_desc()

        self._dump_debug_info(program=program, trainer=trainer)

        trainer_instance = self._default_executor.init_for_dataset(
            program.desc, trainer._desc(), scope, None
        )

        # if fetch_handler is not None:
        #    scope0 = trainer_instance.get_worker_scope(0)
        #    fetch_monitor = FetchHandlerMonitor(scope0, fetch_handler)
        #    fetch_monitor.start()
        #    self._default_executor.run_from_dataset(trainer_instance)
        #    fetch_monitor.stop()
        #    self._default_executor.release_trainer(trainer_instance)
        # else:

        self._default_executor.run_from_dataset(trainer_instance)
        # self._default_executor.release_trainer(trainer_instance)

        return trainer_instance

    def train_from_dataset(
        self,