Train on 2 or more GPUs

seungwonpark / melgan

MelGAN vocoder (compatible with NVIDIA/tacotron2)

http://swpark.me/melgan/

BSD 3-Clause "New" or "Revised" License

633 stars 116 forks source link

Train on 2 or more GPUs #27

Closed muranski closed 4 years ago

muranski commented 4 years ago

Awesome work! How should I change this code to train on 2 or more GPUs?

seungwonpark commented 4 years ago

Probably you won't have to. It has been shown that using batch size as 16 leads to best audio fidelity, (see #14) and that only consumes about 4~5GB of GPU memory.