Freddy Cheng

Hi [@moyid](https://github.com/moyid) : do you figure out why you got 0 on BLEU score, I follow the [example](https://github.com/STayinloves/NMT2017-ZH-EN) and also got 0 I am wondering can you share your pre-processing...

Hi @ghrua, Have you found the solution? I am facing the same problem...

Hi @ghrua, Thank you for your suggestions, I will at least try the second one. I have questions regarding the training: 1. Have you tried using fp16 from scratch, will...

Hi @ghrua, I got it, they are using 4096 (tokens)* 8 (GPUs) * 2 (update-freq). Anyways, thank you for your kindly help and suggestions, I really appreciate it. Thank you.

agree @nanmu42 , I want to implement other language, does it have doc or something like it?

Hi @mravanelli,  The scores reported from code with paper are leverageing other resources. While the score I reported here is trained from...

Hi @mravanelli, the score in the table is reported in tst_com.

Hi @TParcollet, I am working on CoVoST2 recipe, once I finish my baseline I will work on it.

HI @mravanelli, @TParcollet, I've updated the results in the table. I also tried to adapt dynamic batch, however, it required more training time than not using it. If I missed...

Hi @TParcollet, I followed [tutorial](https://colab.research.google.com/drive/1mypqbHDrusZaIbqPoiEGY-WIbnpMHa2I?usp=sharing), PR #1174, and [TIMIT seq2seq recipe](https://github.com/speechbrain/speechbrain/blob/develop/recipes/TIMIT/ASR/seq2seq/hparams/train.yaml). Basically, I followed the style from [TIMIT seq2seq recipe](https://github.com/speechbrain/speechbrain/blob/develop/recipes/TIMIT/ASR/seq2seq/hparams/train.yaml) and played a bit parameters, but, it did not work...