hezar.trainer.trainer_utils module

class hezar.trainer.trainer_utils.AverageMeter(name, avg=None, sum=None, count=None, fmt=':f')

Bases: object

Compute and store the average and current value

reset()

update(val, n=1)

class hezar.trainer.trainer_utils.CSVLogger(logs_dir: str, csv_filename: str)

Bases: object

write(logs: dict, step: int)

class hezar.trainer.trainer_utils.MetricsTracker(metrics)

Bases: object

avg()

reset()

update(results)

class hezar.trainer.trainer_utils.TrainerState(epoch: int = 1, total_epochs: int | None = None, global_step: int = 0, epoch_step: int = 0, loss_tracker_sum: float = 0.0, loss_tracker_avg: float = 0.0, metric_for_best_checkpoint: str | None = None, best_metric_value: float | None = None, best_checkpoint: str | None = None, logs_dir: str | None = None)

Bases: object

A Trainer state is a container for holding specific updating values in the training process and is saved when checkpointing.

Parameters:

• epoch – Current epoch number

• total_epochs – Total epochs to train the model

• global_step – Number of the update steps so far, one step is a full training step (one batch)

• epoch_step – Number of the update steps in the current epoch

• loss_tracker_sum – Running sum value of the loss tracker

• loss_tracker_avg – Running mean value of the loss tracker

• metric_for_best_checkpoint – The metric key for choosing the best checkpoint (Also given in the TrainerConfig)

• best_metric_value – The value of the best checkpoint saved so far

• best_checkpoint – Path to the best model checkpoint so far

• logs_dir – Path to the logs directory

best_checkpoint: str = None

best_metric_value: float = None

epoch: int = 1

epoch_step: int = 0

global_step: int = 0

classmethod load(path)

Load a trainer state from path

logs_dir: str = None

loss_tracker_avg: float = 0.0

loss_tracker_sum: float = 0.0

metric_for_best_checkpoint: str = None

save(path, drop_none: bool = False)

Save the state to a .yaml file at path

total_epochs: int = None

update(items: dict, **kwargs)

update_best_results(metric_value, objective, step)

hezar.trainer.trainer_utils.get_distributed_logger(name: str, level: str | None = None, fmt: str | None = None)

Distributed logger is responsible for handling logging on multiple processes/machines

hezar.trainer.trainer_utils.get_lr_scheduler_type(lr_scheduler, schedulers_mapping: dict)

hezar.trainer.trainer_utils.resolve_logdir(log_dir) → str

hezar.trainer.trainer_utils.write_to_tensorboard(writer: SummaryWriter, logs: dict, step: int)