openai

/

whisper-medium.en

like 38

Hi,

I am currently comparing time improvements for openai/whisper-medium.en vs distil-whisper/distil-medium.en

As suggested in the model card for distil-whisper/distil-medium.en, I am using flash attention 2 to get the best results. I don't completely understand the concept behind flash attention 2. Do I need to use it with openai/whisper-medium.en for a fair comparison? If yes,

1. Is it feasible to use flash attention 2 with openai/whisper-medium.en?
2. How?

Thanks!