Would Whisper require M1 GPU support?

[scottleibrand]

To use Whisper for an app like this, I think we’d first want it to add M1 GPU support, as running even the tiny model on CPU is barely above 1x speed for transcription.

openai/whisper#382 isn’t yet merged, and it’s unclear to me what is required to get it working.

Thoughts?

[scottleibrand]

Maybe not, in light of #1 ?

[pavanagrawal123]

@scottleibrand, Yeah, I believe with the smaller models the whisper.cpp implementation is good at running at real-time speed. I ran it on CPU only and got pretty good performance, I believe this is very doable without GPU support.