Whisper-X is completely broken for me after the new update. about whisperx

[OPPEYRADY]

Made a temp fix, by renaming the file "cublas64_11" to cublas64_12".

[StDensity]

any fix other than rename?

[deep-pipeline]

@OPPEYRADY have you managed to find a solution? If so, could you note it and close this issue please (I'm just trying to find ways to reduce live number of open issues to assist developer bandwidth). Thanks.

[OPPEYRADY]

@deep-pipeline
IIRC ctranslate==3.24.0 may have helped, but that was the only solution I found thus far besides renaming the file.

[OPPEYRADY]

unfortunately this needs someone to fix this as its pretty critical but i'm not sure of any real official fixes that can be made besides my hacky methods.

[eashanchawla]

Maybe we can pull an older version of the whisperx repository since this one has a requirement for faster-whisper 1 which inevitably forces us to pull a version of ctranslate that is > 4, which needs CUDA 12. This is rough because Azure ML (where my whisperx was running, doesn't support base images with cuda 12 yet). Another option could be to also use an older base image in my dockerfile and try to update CUDA from there (which can be problematic in itself).