-
Notifications
You must be signed in to change notification settings - Fork 26
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
CUDA support from Windows #62
Comments
Can you try with |
In this case it does compile, and it does generate the DLL
|
Could you please check what's inside the lib folder |
Inside the folder
|
It seems good to me! the Are you able to compile and run |
I am having some trouble with this too. I installed Cuda, set both environment variables:
Then I installed using pip install . my site-packages looks a little different than indicated here, but I see this ggml.dll: Whenever I run my transcription script, my CPU usage shoots up, but my GPU usage does not change, so I am pretty sure that I am not properly utilizing my GPU. Any advice on next things to try? Happy to provide any compilation info, but I do not know how to get it. |
@jts599, if you don't see any GPU usage, it's likely that |
Seems like it should be good?
I certainly had built it once or twice without it installed, but have since reinstalled using: I assume that rebuilds pywhispercpp in the process? I also once removed the entire directory and re-cloned. |
Yeah, seems good. In that case can you run the compilation with the verbose flag, to see what's going on: GGML_CUDA=1 pip install . --verbose |
Yeah, no cuda is there. |
I get an error with WSL about virtualizations not being enabled on my computer that I have spent a number of hours trying to fix already, I have a linux install on my computer too. I also tried there, but am less confident of my CUDA driver config. I may give it another go and see if I can get it to work. |
Yes, if you have a Linux installation, give it a try. It's working on my end. Just make sure the NVIDIA drivers are installed correctly and that nvcc is available system-wide. |
Tried again on my linux install. I see the install log pick up that the environment variable is set, but I am still not seeing it run on my GPU
(Sorry for the formatting. Could not get markdown to cooperate. Figured this is less annoying than a text file) pip install . --verbose(.venv) jts599@jts599-desktop-ubuntu:/media/jts599/Disk5/Transcriber/DigitalJournal2/pywhispercpp$ pip install . --verbose
Successfully built pywhispercpp
Successfully built pywhispercpp |
From the logs, it seems it's not compiling against cuda yet. |
How can I install pywhispercpp with CUDA support from Windows?
Adding the environment variable WHISPER_CUDA=1 does not cause any changes in the compilation
The text was updated successfully, but these errors were encountered: