Faster-Whisper-XXL r192.3.4 won't use GPU, but r192.3 works fine

[limheat]

Faster-Whisper-XXL r192.3.4 won't use GPU. Already tried --device cuda, didn't work.
Faster-Whisper r192.3 works completely fine on GPU.
Btw, is it worth switching to 192.3.4? I prioritize transcription quality over speed.
(Windows)

[Purfview]

won't use GPU. Already tried --device cuda, didn't work.

Post screenshot of error you get, and post the model of your GPU.

[1001ruchka]

Seems like I shouldn't have used quotation marks around the directory. I would never have thought of that, it always used to work when using quotation marks on longer directories like "E:\podcasts".

Yes, it is.
Make it a rule to enclose the file name and path in quotation marks.
And I would like to point out that your command line does not use any of the faster-whisper-xxl features. With this command line you can use the standard version of faster-whisper.
If you don't use --vad_alt_method and/or --ff_mdx_kim2 - it doesn't make sense to use XXL version.

[limheat]

OK, Got it👍

[Purfview]

Path can't end with \ backslash or you'll get syntax error.

[limheat]

That explains a lot. Sadly I was unaware of that and had to learn the hard way...

[Purfview]

Sadly I was unaware of that and had to learn the hard way...

You could learn it soft way if you could articulate your issues two weeks ago.

[iTula]

The same situation for me. And the error is this:
Traceback (most recent call last):
File "D:\whisper-fast_XXL_main_.py", line 1668, in
File "D:\whisper-fast_XXL_main_.py", line 1505, in cli
File "faster_whisper\transcribe.py", line 143, in init
RuntimeError: CUDA failed with error CUDA driver version is insufficient for CUDA runtime version
[42128] Failed to execute script 'main' due to unhandled exception!

[iTula]

I just updated the Nvidia cuda software and it works with cuda.

[Purfview]

XXL is using CUDA12 libs when regular is using CUDA11. You needed to updated video drivers:

https://www.nvidia.com/download/index.aspx?lang=en-us

[1001ruchka]

XXL is using CUDA12 libs when regular is using CUDA11.

I have Visual Studio and CUDA 11 installed on my computer. The paths to CUDA 11 libraries are specified in Windows environment variables. Does it make sense to replace CUDA with version 12 if I plan to use XXL version?

[Purfview]

I think it doesn't matter what libs you have installed as XXL has needed libs in it's folders.