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

[Wilkuuu]

Existing language matches target language
Unable to load any of {libcudnn_ops.so.9.1.0, libcudnn_ops.so.9.1, libcudnn_ops.so.9, libcudnn_ops.so}
Invalid handle. Cannot load symbol cudnnCreateTensorDescriptor
./start.sh: line 8: 8680 Aborted (core dumped) python xtts_demo.py

any ideas? i reinstall cuda, drivers, venv

[rvuyyuru2]

same here

[rvuyyuru2]

https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Debian&target_version=11&target_type=deb_network installed cuda setup ?, https://developer.nvidia.com/cudnn-downloads?target_os=Linux&target_arch=x86_64&Distribution=Debian&target_version=12&target_type=deb_network

[Wilkuuu]

yes, same issue

[Wilkuuu]

i even try rollback to previous version but ... nope...

[rvuyyuru2]

it worked for me

[Wilkuuu]

What works?

[lainedfles]

There's a deep TTS module dependency chain (pytorch) that is limited to cudnn 8. Sadly, the upstream project has ceased development because Coqui has shut down.

Until there's a better option (https://github.com/idiap/coqui-ai-TTS looks promising), your best option is to use docker or podman.

[mommalicious]

same problem. Tho it says 'Warning, existing language does not match target language. Updated lang.txt with target language.' on first line.

Is there really no workarounds here?

@rvuyyuru2 What worked for you exactly? Did you add anything to the code?

---

Edit: Beware, I am not a coder, but I think @rvuyyuru2 means to add this lines in the beginning of the code:
!sudo apt-get install -y nvidia-open
!sudo apt-get -y install cudnn-cuda-12

also changed pydantic to (according to the error):
!pip install pydantic==2.7.4

The code will stop at the list of country of origin and you have to enter the number you want. I used English (UK), which is 31. Then enter. Next is pick the keyboard, I chosed 1. Then wait for the code to finish there.

It worked for me so far.