search

modelscope / FunASR

A Fundamental End-to-End Speech Recognition Toolkit and Open Source SOTA Pretrained Models, Supporting Speech Recognition, Voice Activity Detection, Text Post-processing etc.
https://www.funasr.com
Other
6.45k stars 687 forks source link

微调paraformer-large模型速度慢，更改batchsize后也无效 #783

Closed sunnymoon155 closed 1 year ago

sunnymoon155 commented 1 year ago

4卡训练显存占用只有5G，请问这个怎么能调一下batch size呢？batch_bins这个参数指的是batch size吗，调整后无效呀。 另外数据读取支持wenet类似的tar格式吗

hnluo commented 1 year ago

Please refer to https://alibaba-damo-academy.github.io/FunASR/en/modelscope_pipeline/asr_pipeline.html#finetune-with-pipeline

yangppde commented 1 month ago

请问问题后面有解决吗