AssertionError: Torch not compiled with CUDA enabled

[fznx922]

Hey mate, after installing the newest version fresh im getting this error

File "C:\Users\fznx\anaconda3\envs\whisperxgui\lib\site-packages\torch\cuda__init__.py", line 293, in _lazy_init raise AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled

i havent had any issues running any other cuda related things on my computer, ive double checked and tried to fresh install a few times without any help

any advise or know how to fix it?

thanks :)

[Pikurrot]

Hey, it seems a problem with PyTorch installation to work with CUDA. I assume you have a Windows and GPU in your computer. If you have already installed a CUDA version ≥12, then try uninstalling PyTorch from your environment and installing it again correctly:

conda activate whisperxgui
conda uninstall pytorch torchaudio pytorch-cuda
conda install pytorch torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia

Let me know if you have further problems.

[fznx922]

hey mate, thanks for the reply, after doing that i got this now

ModuleNotFoundError: No module named 'idna'

ive tried manually installing it in the enviroment, it says its installed but doesnt seem to find it regardless

[Pikurrot]

This is strange. Make sure in your config.json file inside configs/ you have a variable "env_name" with the exact same name as your conda environment, which should be whisperxgui. This is what the script uses to run the program, but you can try to run it yourself manually:

conda activate whisperxgui
python main.py --autolaunch

This is assuming you've installed "idna" correctly with pip install idna. However, it's strange because this should have been installed automatically.

[fznx922]

so i tried a complete fresh environment, now could get back in, but still having the torch compile error again, this is with cuda toolkit installed really scratching my head, cant figure out whats doing it

did notice this at the start Model was trained with torch 1.10.0+cu102, yours is 2.2.0+cpu.

but definitionally chose GPU it detected it fine in the pre installation trying to uninstall and reinstall caused the idna issue again so i guess when it uninstalled it took something with it that it didnt get back reinstalling torch

[Pikurrot]

Don't worry about the Model was trained ... warning, that often appears in WhisperX and is not a problem. Regarding your error, CUDA errors are popular and often really diverse across different computers or OS. I couldn't replicate your problem so I'm not sure how to help you. :disappointed: Again, this is a CUDA error, shouldn't have nothing to do at all with my program, so you could try to rerun the whisper-gui.bat script, creating a new environment but this time selecting to not use the GPU. Note that you would have to delete the config.json file before, for a fresh start. So with CPU should work properly.