From 6f7e27eb7c2d0a7649ec8f14d167c8da8e29f906 Mon Sep 17 00:00:00 2001
From: jmwang66 <wangjiaming.wjm@alibaba-inc.com>
Date: 星期二, 16 五月 2023 15:07:20 +0800
Subject: [PATCH] Merge pull request #518 from alibaba-damo-academy/dev_wjm2
---
funasr/tasks/abs_task.py | 55 ++++++++++++++++++++++++++++++++-----------------------
1 files changed, 32 insertions(+), 23 deletions(-)
diff --git a/funasr/tasks/abs_task.py b/funasr/tasks/abs_task.py
index 55a5d79..361ff89 100644
--- a/funasr/tasks/abs_task.py
+++ b/funasr/tasks/abs_task.py
@@ -30,6 +30,7 @@
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
@@ -44,19 +45,18 @@
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
@@ -230,8 +230,8 @@
>>> 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 +251,8 @@
>>> 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
@@ -263,7 +263,7 @@
@classmethod
@abstractmethod
- def build_model(cls, args: argparse.Namespace) -> AbsESPnetModel:
+ def build_model(cls, args: argparse.Namespace) -> FunASRModel:
raise NotImplementedError
@classmethod
@@ -1172,7 +1172,8 @@
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 +1182,10 @@
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")]
@@ -1244,9 +1247,9 @@
# 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),
@@ -1374,15 +1377,21 @@
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,
+ 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,
+ 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")
elif args.dataset_type == "small":
@@ -1929,7 +1938,7 @@
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.
@@ -1956,9 +1965,9 @@
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:
--
Gitblit v1.9.1