libcudnn_cnn problem

[781574155]

INFO:faster_whisper:Processing audio with duration 03:52.176
Unable to load any of {libcudnn_cnn.so.9.1.0, libcudnn_cnn.so.9.1, libcudnn_cnn.so.9, libcudnn_cnn.so}
Invalid handle. Cannot load symbol cudnnCreateConvolutionDescriptor

ctranslate2==4.5.0
torch==2.5.1
faster-whisper==1.1.0

[781574155]

after setup LD_LIBRARY_PATH, program run correctly.

export LD_LIBRARY_PATH=/usr/local/lib/python3.11/site-packages/nvidia/cudnn/lib/:$LD_LIBRARY_PATH

It's should be correctly setup automatically!

[likeleto]

Same thing via google colab
Unable to load any of {libcudnn_ops.so.9.1.0, libcudnn_ops.so.9.1, libcudnn_ops.so.9, libcudnn_ops.so}

!pip install ctranslate2==4.4.0 helps, but not sure 100%

[umitgunduz]

after setup LD_LIBRARY_PATH, program run correctly.

export LD_LIBRARY_PATH=/usr/local/lib/python3.11/site-packages/nvidia/cudnn/lib/:$LD_LIBRARY_PATH

It's should be correctly setup automatically!

If you are using a virtual environment, following the example below may solve your problem.

export LD_LIBRARY_PATH=.venv/lib/python3.11/site-packages/nvidia/cudnn/lib/:$LD_LIBRARY_PATH

If you are using it as a Docker image...

ENV LD_LIBRARY_PATH=.venv/lib/python3.11/site-packages/nvidia/cudnn/lib

[wredan]

I ran into the same issue:

Unable to load any of {libcudnn_ops.so.9.1.0, libcudnn_ops.so.9.1, libcudnn_ops.so.9, libcudnn_ops.so}

Running the following command solved it for me:

conda install -c nvidia/label/cuda-12.4.0 cudnn

My setup:

• faster-whisper: 1.1.1
• ctranslate2: 4.5.0
• PyTorch: 2.5.1
• CUDA: 12.4
• cuDNN: 9.1.1.17 was missing or not properly installed

Installing cuDNN via Conda resolved the problem.

Update:

If you are using faster-whisper with docker, using this docker image from nvidia solved my issue:

FROM nvidia/cuda:12.4.1-cudnn-runtime-ubuntu22.04

It has the right version of cudnn pre compiled for the cuda version I am using. Check for your version here.

Update 2:

If you are using conda envs, sometimes torch is not loading the right version of the cuDNN, giving:

cuDNN failed with status CUDNN_STATUS_SUBLIBRARY_VERSION_MISMATCH

what solved for me is:

export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"

so that you force torch to load the right cuDNN version.