why the speaker model can not train by multi gpus?

clovaai / voxceleb_trainer

In defence of metric learning for speaker recognition

MIT License

1.03k stars 272 forks source link

why the speaker model can not train by multi gpus? #42

Closed zh794390558 closed 4 years ago

joonson commented 4 years ago

We have found that multi-GPU training does not match the performance on single GPU training. If you would still like to try, there is a pending pull request from another user.