update model-revision by model name

雾聪

2023-12-04 ef07c0d48faa0c19d4ac35f23832069c1bfe04bf

 funasr/train/trainer.py

@@ -26,7 +26,6 @@
import torch
import torch.nn
import torch.optim
from typeguard import check_argument_types

from funasr.iterators.abs_iter_factory import AbsIterFactory
from funasr.main_funcs.average_nbest_models import average_nbest_models
@@ -127,7 +126,6 @@
    @classmethod
    def build_options(cls, args: argparse.Namespace) -> TrainerOptions:
        """Build options consumed by train(), eval()"""
        assert check_argument_types()
        return build_dataclass(TrainerOptions, args)

    @classmethod
@@ -188,7 +186,6 @@
        distributed_option: DistributedOption,
    ) -> None:
        """Perform training. This method performs the main process of training."""
        assert check_argument_types()
        # NOTE(kamo): Don't check the type more strictly as far trainer_options
        assert is_dataclass(trainer_options), type(trainer_options)
        assert len(optimizers) == len(schedulers), (len(optimizers), len(schedulers))
@@ -281,14 +278,11 @@
        for iepoch in range(start_epoch, trainer_options.max_epoch + 1):
            if iepoch != start_epoch:
                logging.info(
                    "{}/{}epoch started. Estimated time to finish: {}".format(
                    "{}/{}epoch started. Estimated time to finish: {} hours".format(
                        iepoch,
                        trainer_options.max_epoch,
                        humanfriendly.format_timespan(
                            (time.perf_counter() - start_time)
                            / (iepoch - start_epoch)
                            * (trainer_options.max_epoch - iepoch + 1)
                        ),
                        (time.perf_counter() - start_time) / 3600.0 / (iepoch - start_epoch) * (
                                trainer_options.max_epoch - iepoch + 1),
                    )
                )
            else:
@@ -372,7 +366,7 @@
                            ],
                            "scaler": scaler.state_dict() if scaler is not None else None,
                            "ema_model": model.encoder.ema.model.state_dict()
                            if hasattr(model.encoder, "ema") and model.encoder.ema is not None else None,
                            if hasattr(model, "encoder") and hasattr(model.encoder, "ema") and model.encoder.ema is not None else None,
                        },
                        buffer,
                    )
@@ -551,7 +545,6 @@
        options: TrainerOptions,
        distributed_option: DistributedOption,
    ) -> Tuple[bool, bool]:
        assert check_argument_types()

        grad_noise = options.grad_noise
        accum_grad = options.accum_grad
@@ -845,7 +838,6 @@
        options: TrainerOptions,
        distributed_option: DistributedOption,
    ) -> None:
        assert check_argument_types()
        ngpu = options.ngpu
        no_forward_run = options.no_forward_run
        distributed = distributed_option.distributed