speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

update trainer

Browse Source
This commit is contained in:
游雁 2024-02-21 19:22:01 +08:00
parent b2ea4424a3
commit 2bc330a599
5 changed files with 19 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

0
examples/industrial_data_pretraining/paraformer/infer_demo.sh → examples/industrial_data_pretraining/paraformer/demo.sh
View File

2
examples/industrial_data_pretraining/paraformer/finetune.sh
View File

@ -6,7 +6,7 @@
#git clone https://www.modelscope.cn/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch.git ${local_path} #git clone https://www.modelscope.cn/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch.git ${local_path}
## generate jsonl from wav.scp and text.txt ## generate jsonl from wav.scp and text.txt
#python funasr/datasets/audio_datasets/scp2jsonl.py \ #python -m funasr.datasets.audio_datasets.scp2jsonl \
#++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \ #++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \
#++data_type_list='["source", "target"]' \ #++data_type_list='["source", "target"]' \
#++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl #++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl

16
funasr/datasets/audio_datasets/scp2jsonl.py
View File

@ -72,14 +72,7 @@ def parse_context_length(data_list: list, data_type: str):
@hydra.main(config_name=None, version_base=None) @hydra.main(config_name=None, version_base=None)
def main_hydra(cfg: DictConfig): def main_hydra(cfg: DictConfig):
"""
python funasr/datasets/audio_datasets/scp2jsonl.py \
++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \
++data_type_list='["source", "target"]' \
++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl
"""
kwargs = OmegaConf.to_container(cfg, resolve=True) kwargs = OmegaConf.to_container(cfg, resolve=True)
scp_file_list = kwargs.get("scp_file_list", ("/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt")) scp_file_list = kwargs.get("scp_file_list", ("/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"))
@ -90,6 +83,13 @@ def main_hydra(cfg: DictConfig):
gen_jsonl_from_wav_text_list(scp_file_list, data_type_list=data_type_list, jsonl_file_out=jsonl_file_out) gen_jsonl_from_wav_text_list(scp_file_list, data_type_list=data_type_list, jsonl_file_out=jsonl_file_out)
"""
python -m funasr.datasets.audio_datasets.scp2jsonl \
++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \
++data_type_list='["source", "target"]' \
++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl
"""
if __name__ == "__main__": if __name__ == "__main__":
main_hydra() main_hydra()

2
funasr/models/mossformer/mossformer_encoder.py
View File

@ -4,7 +4,7 @@ import torch.nn.functional as F
try: try:
from rotary_embedding_torch import RotaryEmbedding from rotary_embedding_torch import RotaryEmbedding
except: except:
print("Please install rotary_embedding_torch by: \n pip install -U rotary_embedding_torch") print("If you want use mossformer, lease install rotary_embedding_torch by: \n pip install -U rotary_embedding_torch")
from funasr.models.transformer.layer_norm import GlobalLayerNorm, CumulativeLayerNorm, ScaleNorm from funasr.models.transformer.layer_norm import GlobalLayerNorm, CumulativeLayerNorm, ScaleNorm
from funasr.models.transformer.embedding import ScaledSinuEmbedding from funasr.models.transformer.embedding import ScaledSinuEmbedding
from funasr.models.transformer.mossformer import FLASH_ShareA_FFConvM from funasr.models.transformer.mossformer import FLASH_ShareA_FFConvM

18
funasr/train_utils/trainer.py
View File

@ -302,17 +302,14 @@ class Trainer:
) )
pbar.set_description(description) pbar.set_description(description)
if self.writer: if self.writer:
self.writer.add_scalar(f'rank{self.local_rank}_Loss/train', loss.item(), self.writer.add_scalar(f'rank{self.local_rank}_Loss/train', loss.item(), self.batch_total)
epoch*len(self.dataloader_train) + batch_idx) self.writer.add_scalar(f'rank{self.local_rank}_lr/train', lr, self.batch_total)
for key, var in stats.items(): for key, var in stats.items():
self.writer.add_scalar(f'rank{self.local_rank}_{key}/train', var.item(), self.writer.add_scalar(f'rank{self.local_rank}_{key}/train', var.item(), self.batch_total)
epoch * len(self.dataloader_train) + batch_idx)
for key, var in speed_stats.items(): for key, var in speed_stats.items():
self.writer.add_scalar(f'rank{self.local_rank}_{key}/train', eval(var), self.writer.add_scalar(f'rank{self.local_rank}_{key}/train', eval(var), self.batch_total)
epoch * len(self.dataloader_train) + batch_idx)
# if batch_idx == 2:
# break
pbar.close() pbar.close()
def _validate_epoch(self, epoch): def _validate_epoch(self, epoch):
@ -356,7 +353,10 @@ class Trainer:
if (batch_idx+1) % self.log_interval == 0 or (batch_idx+1) == len(self.dataloader_val): if (batch_idx+1) % self.log_interval == 0 or (batch_idx+1) == len(self.dataloader_val):
pbar.update(self.log_interval) pbar.update(self.log_interval)
time_now = datetime.now()
time_now = time_now.strftime("%Y-%m-%d %H:%M:%S")
description = ( description = (
f"{time_now}, "
f"rank: {self.local_rank}, " f"rank: {self.local_rank}, "
f"validation epoch: {epoch}/{self.max_epoch}, " f"validation epoch: {epoch}/{self.max_epoch}, "
f"step: {batch_idx+1}/{len(self.dataloader_val)}, " f"step: {batch_idx+1}/{len(self.dataloader_val)}, "