Wenxin comments

Results 5 comments of


                                            Wenxin

Support cpu/gpu utterance batch decoding for Transformer ASR model

![图片](https://user-images.githubusercontent.com/22743466/103450455-b0161a80-4cf1-11eb-92ba-9baddd2b74fd.png)

Support cpu/gpu utterance batch decoding for Transformer ASR model

> > > Can you add tests comparing `recognize_batch` vs for-loop `recognize` results like this? > > https://github.com/espnet/espnet/blob/b6f7e53ade7206c33bc9beac62cdc191a0c761be/test/test_batch_beam_search.py#L89 No problem. But since I have no previous experience in using pytest,...

when I execute the command "python train.py --actor-model facebook/opt-1.3b --reward-model facebook/opt-350m --deployment-type single_gpu" I got the follow result, and I try many solutions but all failure.

same question here

About development set and the number of training data

Thank you for your response! I am willing to help but I have no experience in splitting valid data in ASR... About the data amount, I am looking forward to...

AttributeError: module 'lightseq' has no attribute 'training'

Same problem here.