SystemResourcesMonitor

class torchtnt.framework.callbacks.SystemResourcesMonitor(loggers: Union[MetricLogger, List[MetricLogger]], *, logging_interval: Literal['epoch', 'step'] = 'epoch')

A callback which logs system stats, including: - CPU usage - resident set size - GPU usage - cuda memory stats

Parameters:

• loggers – Either a torchtnt.loggers.logger.MetricLogger or list of torchtnt.loggers.logger.MetricLogger
• logging_interval – whether to print system state every step or every epoch. Defaults to every epoch.

on_eval_epoch_start(state: State, unit: EvalUnit[TEvalData]) → None

Hook called before a new eval epoch starts.

on_eval_step_start(state: State, unit: EvalUnit[TEvalData]) → None

Hook called before a new eval step starts.

on_predict_epoch_start(state: State, unit: PredictUnit[TPredictData]) → None

Hook called before a new predict epoch starts.

on_predict_step_start(state: State, unit: PredictUnit[TPredictData]) → None

Hook called before a new predict step starts.

on_train_epoch_start(state: State, unit: TrainUnit[TTrainData]) → None

Hook called before a new train epoch starts.

on_train_step_start(state: State, unit: TrainUnit[TTrainData]) → None

Hook called before a new train step starts.