libcudnn - Error

[Q-si]

When I try to perform a transcription using the Swagger Page API, I get the following error:

Could not load library libcudnn_ops_infer.so.8. Error: libcudnn_ops_infer.so.8: cannot open shared object file: No such file or directory

Any idea where the problem is and how to solve it?

[jhj0517]

Trying to update faster-whisper and torch might help:

pip install -U faster-whisper torch torchaudio --extra-index-url https://download.pytorch.org/whl/cu124

( This assumes that you're using CUDA 12.4 ),
This happens because there's a strict version matrix between CTranslate2 and torch : SYSTRAN/faster-whisper#1086

[Q-si]

This solved the issue :)
But now, i've got another error: TranscriptionOptions.__new__() missing 1 required positional argument: 'hotwords'
It appears that there is a new parameter "hotwords" that should be added to the Swagger page.

[Q-si]

nvidia-smi is showing me Driver Version: 550.120 and CUDA Version: 12.4 . But i am not able to build a Docker Container that works :(
Can someone push a Container to hub.docker.com?

[Q-si]

I've added "ctranslate2==4.4.0" at the end of the file requirements.txt
It seems to work now.