Add Accelerate-enabled trainer

lucidrains / voicebox-pytorch

Implementation of Voicebox, new SOTA Text-to-speech network from MetaAI, in Pytorch

MIT License

589 stars 49 forks source link

Add Accelerate-enabled trainer #27

Closed lucasnewman closed 11 months ago

lucasnewman commented 11 months ago

This adds a simple Accelerate-enabled trainer class that can be used to train against audio-only data, optionally conditioned on the semantic tokens coming from the Spear-TTS wav2vec implementation. I verified the loss converges on LibriTTS-R.

lucidrains commented 11 months ago

boss