Jacob Andersson

@sssshhhhhh yes might just be BLAS shortcuts and I would expect some instabilities, but 1e-4 feels like to large of a diff. I have compared the weights that are loaded...

yes for a single layer this might be fine. But the error enlarges as we go through the encoder so the final encoded features are quite off. The start of...

@sssshhhhhh sorry for the very late response. Could you please share your code so that I can see if I am doing something wrong or different on my end. When...

@nullscc Did you find the problem? I am experiencing the exact same issue when i compare performance of a finetuned whisper model between faster whisper and the original whisper. Looking...