mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-09-03 14:14:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fixes after comment.

Browse Source
This commit is contained in:
Fangjun Kuang 2022-03-31 20:29:40 +08:00
parent 9e5ca821bc
commit 4f87802200
2 changed files with 20 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
egs/librispeech/ASR/tdnn_lstm_ctc/asr_datamodule.py
View File

@ -42,11 +42,11 @@ from icefall.utils import str2bool
class _SeedWorkers:
def __init__(self, seed: int):
self.seed = seed
def __call__(self, worker_id: int):
# 'seed' is derived from the current random state, which will have
# previously been set in the main process.
seed = torch.randint(0, 100000, ()).item()
fix_random_seed(seed + worker_id)
fix_random_seed(self.seed + worker_id)
class LibriSpeechAsrDataModule:
@ -311,13 +311,18 @@ class LibriSpeechAsrDataModule:
logging.info("Loading sampler state dict")
train_sampler.load_state_dict(sampler_state_dict)
# 'seed' is derived from the current random state, which will have
# previously been set in the main process.
seed = torch.randint(0, 100000, ()).item()
worker_init_fn = _SeedWorkers(seed)
train_dl = DataLoader(
train,
sampler=train_sampler,
batch_size=None,
num_workers=self.args.num_workers,
persistent_workers=False,
worker_init_fn=_SeedWorkers(),
worker_init_fn=worker_init_fn,
)
return train_dl

15
egs/librispeech/ASR/transducer_stateless_multi_datasets/asr_datamodule.py
View File

@ -42,11 +42,11 @@ from icefall.utils import str2bool
class _SeedWorkers:
def __init__(self, seed: int):
self.seed = seed
def __call__(self, worker_id: int):
# 'seed' is derived from the current random state, which will have
# previously been set in the main process.
seed = torch.randint(0, 100000, ()).item()
fix_random_seed(seed + worker_id)
fix_random_seed(self.seed + worker_id)
class AsrDataModule:
@ -264,13 +264,18 @@ class AsrDataModule:
logging.info("About to create train dataloader")
# 'seed' is derived from the current random state, which will have
# previously been set in the main process.
seed = torch.randint(0, 100000, ()).item()
worker_init_fn = _SeedWorkers(seed)
train_dl = DataLoader(
train,
sampler=train_sampler,
batch_size=None,
num_workers=self.args.num_workers,
persistent_workers=False,
worker_init_fn=_SeedWorkers(),
worker_init_fn=worker_init_fn,
)
return train_dl