search

lucidrains / audiolm-pytorch

Implementation of AudioLM, a SOTA Language Modeling Approach to Audio Generation out of Google Research, in Pytorch
MIT License
2.36k stars 255 forks source link

Option to pass in Accelerator to SoundStreamTrainer #142

Closed hmartiro closed 1 year ago

hmartiro commented 1 year ago

This is needed for multi-GPU use cases where certain actions need to be performed on the main process before the trainer is constructed.

Also minor tweak to accept more audio types in the data loader.

lucidrains commented 1 year ago

@hmartiro thank you Hayk! yes, this extra bit of flexibility makes sense