Python trainer.Trainer方法代码示例

# 需要导入模块: from fairseq import trainer [as 别名]
# 或者: from fairseq.trainer import Trainer [as 别名]
def get_parser_with_args(default_task="pytorch_translate"):
    parser = options.get_parser("Trainer", default_task=default_task)
    pytorch_translate_options.add_verbosity_args(parser, train=True)
    pytorch_translate_options.add_dataset_args(parser, train=True, gen=True)
    options.add_distributed_training_args(parser)
    # Adds args related to training (validation and stopping criterions).
    optimization_group = options.add_optimization_args(parser)
    pytorch_translate_options.expand_optimization_args(optimization_group)
    # Adds args related to checkpointing.
    checkpointing_group = options.add_checkpoint_args(parser)
    pytorch_translate_options.expand_checkpointing_args(checkpointing_group)
    # Add model related args
    options.add_model_args(parser)
    # Adds args for generating intermediate BLEU eval while training.
    generation_group = options.add_generation_args(parser)
    pytorch_translate_options.expand_generation_args(generation_group, train=True)
    # Adds args related to input data files (preprocessing, numberizing, and
    # binarizing text files; creating vocab files)
    pytorch_translate_options.add_preprocessing_args(parser)
    return parser 

# 需要导入模块: from fairseq import trainer [as 别名]
# 或者: from fairseq.trainer import Trainer [as 别名]
def single_process_main(args, trainer_class=Trainer, **train_step_kwargs):
    """Train the model for multiple epochs."""
    pytorch_translate_options.print_args(args)
    trainer, task, epoch_itr = setup_training(args, trainer_class)
    extra_state, epoch_itr, checkpoint_manager = setup_training_state(
        args=args, trainer=trainer, task=task, epoch_itr=epoch_itr
    )
    train(
        args=args,
        extra_state=extra_state,
        trainer=trainer,
        task=task,
        epoch_itr=epoch_itr,
        checkpoint_manager=checkpoint_manager,
        **train_step_kwargs,
    ) 

# 需要导入模块: from fairseq import trainer [as 别名]
# 或者: from fairseq.trainer import Trainer [as 别名]
def main(args, trainer_class=Trainer, **train_step_kwargs):
    # We preprocess the data (generating vocab files and binarized data files
    # if needed) outside of the train processes to prevent them from having to
    # wait while the master process is doing this.
    preprocess.preprocess_corpora(args)

    if args.distributed_world_size == 1:
        single_process_main(args, trainer_class, **train_step_kwargs)
    else:
        spawn_context, output_queue = multi_process_main(args=args, start_rank=0)

        while not spawn_context.join(timeout=30):
            # Periodically clears the output queue to ensure that the processes
            # don't deadlock due to queue buffer being full. This is also
            # necessary to ensure that processes join correctly, since a process
            # may not terminate until all items it put on the queue have been
            # consumed (per
            # https://docs.python.org/3/library/multiprocessing.html#all-start-methods).
            try:
                while True:
                    output_queue.get_nowait()
            except queue.Empty:
                pass 

# 需要导入模块: from fairseq import trainer [as 别名]
# 或者: from fairseq.trainer import Trainer [as 别名]
def gpu_train_step(test_args: ModelParamsDict) -> Tuple[Trainer, Dict[Any, Any]]:
    """Sets up inputs from test_args then executes a single train step. A train
    step always requires a GPU."""
    samples, src_dict, tgt_dict = prepare_inputs(test_args)
    task = tasks.DictionaryHolderTask(src_dict, tgt_dict)
    model = task.build_model(test_args)
    criterion = task.build_criterion(test_args)
    sample = next(samples)
    trainer = Trainer(test_args, task, model, criterion)
    logging_dict = trainer.train_step([sample])
    return trainer, logging_dict 

# 需要导入模块: from fairseq import trainer [as 别名]
# 或者: from fairseq.trainer import Trainer [as 别名]
def setup_training(args, trainer_class=None):
    """ Perform several steps:
    - build model using provided criterion and task
    - load data
    - build trainer
    """

    # Overrides the default print() to always prepend the timestamp for more
    # informative logging.
    builtin_print = __builtin__.print

    def print(*args, **kwargs):
        if "file" not in kwargs:
            builtin_print(
                f"[{datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S.%f')}]",
                *args,
                **kwargs,
            )
        else:
            builtin_print(*args, **kwargs)

    __builtin__.print = print

    task, model, criterion = setup_training_model(args)
    if trainer_class is None:
        trainer_class = Trainer

    trainer, epoch_itr = build_trainer(
        args=args,
        task=task,
        model=model,
        criterion=criterion,
        trainer_class=trainer_class,
    )

    return trainer, task, epoch_itr