search

huggingface / distil-whisper

Distilled variant of Whisper for speech recognition. 6x faster, 50% smaller, within 1% word error rate.
MIT License
3.33k stars 238 forks source link

Accelerate #59

Open wntg opened 6 months ago

wntg commented 6 months ago

Accelerate is a tool for multi-machine，but why you use it in single gpu？

sanchit-gandhi commented 6 months ago

Accelerate makes it trivial to scale from a single machine to multiple machines with the same code. In this regard, the training code is generalisable, meaning the same training loop can be used agnostic of the machine https://huggingface.co/docs/accelerate/basic_tutorials/launch

Therefore, users can leverage whatever compute they have available, with no code changes to the distillation code