funasr pages

游雁

2023-11-16 4ace5a95b052d338947fc88809a440ccd55cf6b4

 funasr/tasks/abs_task.py

@@ -30,9 +30,8 @@
import torch.nn
import torch.optim
import yaml
from funasr.models.base_model import FunASRModel
from torch.utils.data import DataLoader
from typeguard import check_argument_types
from typeguard import check_return_type

from funasr import __version__
from funasr.datasets.dataset import AbsDataset
@@ -44,19 +43,18 @@
from funasr.iterators.multiple_iter_factory import MultipleIterFactory
from funasr.iterators.sequence_iter_factory import SequenceIterFactory
from funasr.main_funcs.collect_stats import collect_stats
from funasr.optimizers.sgd import SGD
from funasr.optimizers.fairseq_adam import FairseqAdam
from funasr.optimizers.sgd import SGD
from funasr.samplers.build_batch_sampler import BATCH_TYPES
from funasr.samplers.build_batch_sampler import build_batch_sampler
from funasr.samplers.unsorted_batch_sampler import UnsortedBatchSampler
from funasr.schedulers.noam_lr import NoamLR
from funasr.schedulers.warmup_lr import WarmupLR
from funasr.schedulers.tri_stage_scheduler import TriStageLR
from funasr.schedulers.warmup_lr import WarmupLR
from funasr.torch_utils.load_pretrained_model import load_pretrained_model
from funasr.torch_utils.model_summary import model_summary
from funasr.torch_utils.pytorch_version import pytorch_cudnn_version
from funasr.torch_utils.set_all_random_seed import set_all_random_seed
from funasr.train.abs_espnet_model import AbsESPnetModel
from funasr.train.class_choices import ClassChoices
from funasr.train.distributed_utils import DistributedOption
from funasr.train.trainer import Trainer
@@ -73,6 +71,7 @@
from funasr.utils.types import str_or_none
from funasr.utils.wav_utils import calc_shape, generate_data_list, filter_wav_text
from funasr.utils.yaml_no_alias_safe_dump import yaml_no_alias_safe_dump
from funasr.modules.lora.utils import mark_only_lora_as_trainable

try:
    import wandb
@@ -230,8 +229,8 @@
        >>> cls.check_task_requirements()
        If your model is defined as following,

        >>> from funasr.train.abs_espnet_model import AbsESPnetModel
        >>> class Model(AbsESPnetModel):
        >>> from funasr.models.base_model import FunASRModel
        >>> class Model(FunASRModel):
        ...     def forward(self, input, output, opt=None):  pass

        then "required_data_names" should be as
@@ -251,8 +250,8 @@
        >>> cls.check_task_requirements()
        If your model is defined as follows,

        >>> from funasr.train.abs_espnet_model import AbsESPnetModel
        >>> class Model(AbsESPnetModel):
        >>> from funasr.models.base_model import FunASRModel
        >>> class Model(FunASRModel):
        ...     def forward(self, input, output, opt=None):  pass

        then "optional_data_names" should be as
@@ -263,12 +262,12 @@

    @classmethod
    @abstractmethod
    def build_model(cls, args: argparse.Namespace) -> AbsESPnetModel:
    def build_model(cls, args: argparse.Namespace) -> FunASRModel:
        raise NotImplementedError


    @classmethod
    def get_parser(cls) -> config_argparse.ArgumentParser:
        assert check_argument_types()

        class ArgumentDefaultsRawTextHelpFormatter(
            argparse.RawTextHelpFormatter,
@@ -954,11 +953,22 @@
            default=None,
            help="oss bucket.",
        )
        group.add_argument(
            "--enable_lora",
            type=str2bool,
            default=False,
            help="Apply lora for finetuning.",
        )
        group.add_argument(
            "--lora_bias",
            type=str,
            default="none",
            help="lora bias.",
        )

        cls.trainer.add_arguments(parser)
        cls.add_task_arguments(parser)

        assert check_return_type(parser)
        return parser

    @classmethod
@@ -1006,7 +1016,6 @@
            return _cls

        # This method is used only for --print_config
        assert check_argument_types()
        parser = cls.get_parser()
        args, _ = parser.parse_known_args()
        config = vars(args)
@@ -1046,7 +1055,6 @@

    @classmethod
    def check_required_command_args(cls, args: argparse.Namespace):
        assert check_argument_types()
        if hasattr(args, "required"):
            for k in vars(args):
                if "-" in k:
@@ -1076,7 +1084,6 @@
            inference: bool = False,
    ) -> None:
        """Check if the dataset satisfy the requirement of current Task"""
        assert check_argument_types()
        mes = (
            f"If you intend to use an additional input, modify "
            f'"{cls.__name__}.required_data_names()" or '
@@ -1103,14 +1110,12 @@

    @classmethod
    def print_config(cls, file=sys.stdout) -> None:
        assert check_argument_types()
        # Shows the config: e.g. python train.py asr --print_config
        config = cls.get_default_config()
        file.write(yaml_no_alias_safe_dump(config, indent=4, sort_keys=False))

    @classmethod
    def main(cls, args: argparse.Namespace = None, cmd: Sequence[str] = None):
        assert check_argument_types()
        print(get_commandline_args(), file=sys.stderr)
        if args is None:
            parser = cls.get_parser()
@@ -1147,7 +1152,6 @@

    @classmethod
    def main_worker(cls, args: argparse.Namespace):
        assert check_argument_types()

        # 0. Init distributed process
        distributed_option = build_dataclass(DistributedOption, args)
@@ -1172,7 +1176,8 @@
                    args.batch_bins = args.batch_bins * args.ngpu

        # filter samples if wav.scp and text are mismatch
        if (args.train_shape_file is None and args.dataset_type == "small") or args.train_data_file is None and args.dataset_type == "large":
        if (
                args.train_shape_file is None and args.dataset_type == "small") or args.train_data_file is None and args.dataset_type == "large":
            if not args.simple_ddp or distributed_option.dist_rank == 0:
                filter_wav_text(args.data_dir, args.train_set)
                filter_wav_text(args.data_dir, args.dev_set)
@@ -1181,8 +1186,10 @@

        if args.train_shape_file is None and args.dataset_type == "small":
            if not args.simple_ddp or distributed_option.dist_rank == 0:
                calc_shape(args.data_dir, args.train_set, args.frontend_conf, args.speech_length_min, args.speech_length_max)
                calc_shape(args.data_dir, args.dev_set, args.frontend_conf, args.speech_length_min, args.speech_length_max)
                calc_shape(args.data_dir, args.train_set, args.frontend_conf, args.speech_length_min,
                           args.speech_length_max)
                calc_shape(args.data_dir, args.dev_set, args.frontend_conf, args.speech_length_min,
                           args.speech_length_max)
            if args.simple_ddp:
                dist.barrier()
            args.train_shape_file = [os.path.join(args.data_dir, args.train_set, "speech_shape")]
@@ -1244,14 +1251,16 @@

        # 2. Build model
        model = cls.build_model(args=args)
        if not isinstance(model, AbsESPnetModel):
        if not isinstance(model, FunASRModel):
            raise RuntimeError(
                f"model must inherit {AbsESPnetModel.__name__}, but got {type(model)}"
                f"model must inherit {FunASRModel.__name__}, but got {type(model)}"
            )
        model = model.to(
            dtype=getattr(torch, args.train_dtype),
            device="cuda" if args.ngpu > 0 else "cpu",
        )
        if args.enable_lora:
            mark_only_lora_as_trainable(model, args.lora_bias)
        for t in args.freeze_param:
            for k, p in model.named_parameters():
                if k.startswith(t + ".") or k == t:
@@ -1372,19 +1381,10 @@

            # 7. Build iterator factories
            if args.dataset_type == "large":
                from funasr.datasets.large_datasets.build_dataloader import ArkDataLoader
                train_iter_factory = ArkDataLoader(args.train_data_file, args.token_list, args.dataset_conf,
                                                   frontend_conf=args.frontend_conf if hasattr(args, "frontend_conf") else None,
                                                   seg_dict_file=args.seg_dict_file if hasattr(args, "seg_dict_file") else None,
                                                   punc_dict_file=args.punc_list if hasattr(args, "punc_list") else None,
                                                   bpemodel_file=args.bpemodel if hasattr(args, "bpemodel") else None,
                                                   mode="train")
                valid_iter_factory = ArkDataLoader(args.valid_data_file, args.token_list, args.dataset_conf, 
                                                   frontend_conf=args.frontend_conf if hasattr(args, "frontend_conf") else None,
                                                   seg_dict_file=args.seg_dict_file if hasattr(args, "seg_dict_file") else None,
                                                   punc_dict_file=args.punc_list if hasattr(args, "punc_list") else None,
                                                   bpemodel_file=args.bpemodel if hasattr(args, "bpemodel") else None,
                                                   mode="eval")
                from funasr.datasets.large_datasets.build_dataloader import LargeDataLoader
                train_iter_factory = LargeDataLoader(args, mode="train")
                valid_iter_factory = LargeDataLoader(args, mode="eval")

            elif args.dataset_type == "small":
                train_iter_factory = cls.build_iter_factory(
                    args=args,
@@ -1561,7 +1561,6 @@
        - 4 epoch with "--num_iters_per_epoch" == 4

        """
        assert check_argument_types()
        iter_options = cls.build_iter_options(args, distributed_option, mode)

        # Overwrite iter_options if any kwargs is given
@@ -1594,7 +1593,6 @@
    def build_sequence_iter_factory(
            cls, args: argparse.Namespace, iter_options: IteratorOptions, mode: str
    ) -> AbsIterFactory:
        assert check_argument_types()

        if hasattr(args, "frontend_conf"):
            if args.frontend_conf is not None and "fs" in args.frontend_conf:
@@ -1688,7 +1686,6 @@
            iter_options: IteratorOptions,
            mode: str,
    ) -> AbsIterFactory:
        assert check_argument_types()

        dataset = ESPnetDataset(
            iter_options.data_path_and_name_and_type,
@@ -1793,7 +1790,6 @@
    def build_multiple_iter_factory(
            cls, args: argparse.Namespace, distributed_option: DistributedOption, mode: str
    ):
        assert check_argument_types()
        iter_options = cls.build_iter_options(args, distributed_option, mode)
        assert len(iter_options.data_path_and_name_and_type) > 0, len(
            iter_options.data_path_and_name_and_type
@@ -1890,7 +1886,6 @@
            inference: bool = False,
    ) -> DataLoader:
        """Build DataLoader using iterable dataset"""
        assert check_argument_types()
        # For backward compatibility for pytorch DataLoader
        if collate_fn is not None:
            kwargs = dict(collate_fn=collate_fn)
@@ -1929,7 +1924,7 @@
            model_file: Union[Path, str] = None,
            cmvn_file: Union[Path, str] = None,
            device: str = "cpu",
    ) -> Tuple[AbsESPnetModel, argparse.Namespace]:
    ) -> Tuple[FunASRModel, argparse.Namespace]:
        """Build model from the files.

        This method is used for inference or fine-tuning.
@@ -1940,7 +1935,6 @@
            device: Device type, "cpu", "cuda", or "cuda:N".

        """
        assert check_argument_types()
        if config_file is None:
            assert model_file is not None, (
                "The argument 'model_file' must be provided "
@@ -1956,9 +1950,9 @@
            args["cmvn_file"] = cmvn_file
        args = argparse.Namespace(**args)
        model = cls.build_model(args)
        if not isinstance(model, AbsESPnetModel):
        if not isinstance(model, FunASRModel):
            raise RuntimeError(
                f"model must inherit {AbsESPnetModel.__name__}, but got {type(model)}"
                f"model must inherit {FunASRModel.__name__}, but got {type(model)}"
            )
        model.to(device)
        if model_file is not None: