poor1017
poor1017
Hi @Samuele Cornell, @Cem Subakan I found the reason that makes DDP do not work: [https://github.com/speechbrain/speechbrain/blob/develop/recipes/WHAMandWHAMR/enhancement/dynamic_mixing.py#L216](url) Can you tell me the reason you wrap `train_data` with `torch.utils.data.DataLoader`? Thank you
I encounter the same issue after changing machine. ENVs do not work.
https://github.com/speechbrain/speechbrain/blob/develop/recipes/VoxLingua107/lang_id/train.py You can refer this script. Keep train, valid, test data as small as possible, you should get the bug.
@rwightman Yes, you are right. Maybe we should output a warning for this situation.
Hi @chinhuang007 Sorry for getting back to you late. I tried the verification, and it shows that the onnx has no issues. Here is the onnx model link: [https://www.dropbox.com/s/5hk2wpb56e2kn0m/enc.onnx?dl=0](url) Could...
In my grpc server, I maintain several processors. When a client connect to the server, one processor will be reset and load some data according to the client, then I...
Emm..... I tried 8k, there is also a little gap between my result and the one in README. I think this is not important.
@popcornell Only with Dynamic Mixing, because with DM, you use torch.utils.data.DataLoader to cover sb.dataio.dataset.DynamicItemDataset @ycemsubakan I have not tried WSJ0Mix.
I have encounter this problem: exploded. Have you solved it? Thanks!
eval mode?