Could not load library libcudnn_ops_infer.so.8.

[AitNev]

Which OS are you using?

• OS: Linux

With the default installation bash Install.sh and Only when using the Faster Whisper implementation i get this error message when trying to transcribe with whisper:

venv ./venv/bin/python

Use Faster Whisper implementation
Device "cuda" is detected
Running on local URL:  http://0.0.0.0:8860

To create a public link, set `share=True` in `launch()`.
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
start-webui.sh: line 7: 141135 Aborted                 $PYTHON ./app.py --server_name 0.0.0.0 --server_port 8860
launching the app
DeprecationWarning: 'source deactivate' is deprecated. Use 'conda deactivate'.

• One way to avoid this issue was adding in start-webui.sh the flag --disable_faster_whisper runs slower but it doesn't throw an error

The way it was fixed from this comment where by only doing the second step, adding to requirements.txt

nvidia-cublas-cu11
nvidia-cudnn-cu11

And then exporting their path by adding this line inside start-webui.sh before executing app.py

export LD_LIBRARY_PATH=`$PYTHON -c 'import os; import nvidia.cublas.lib; import nvidia.cudnn.lib; print(os.path.dirname(nvidia.cublas.lib.__file__) + ":" + os.path.dirname(nvidia.cudnn.lib.__file__))'`

With that I was able to get faster-whisper working but just the nvidia libraries weight 1GB, I'm not sure if someone is able to get it running without those libraries.
At least this might help someone who stumbles on this issue.

[AitNev]

Already was this issue 114

[jhj0517]

Hi, Thanks for reporting. It seems this happens on Linux & ubuntu.
Just noting some related issues here:

• 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 #114
• faster-whisper #297
• faster-whisper #729

[jhj0517]

duplicate with #271, if anyone faced the same error refer to #271 (comment)