CUDA provider not Selectable

[MrDreadfall]

Hi Everyone,

For some reason, when I opened the Browser UI under setting, CPU is pre-selected as Provider. There are no other Options. I am have already installed CUDA toolkit and cudnn like the Instruction, it has no effect at all.

Details What OS are you using?

• [ ] Linux
• [ ] Linux in WSL
• [x] Windows
• [ ] Mac

Are you using a GPU?

• [ ] No. CPU FTW
• [x ] NVIDIA
• [ ] AMD
• [ ] Intel
• [ ] Mac

[lysxelapsed]

There still seems to be something missing or not supported. What's your graphics card?

[MrDreadfall]

I'm using a RTX 4070.

[kingmike67]

Any chance you are using the NVidia Studio drivers? I ran into the same issue with my 3060 with Studio Drivers. I switched to Game Ready and its working

[MrDreadfall]

Thanks for the answer! I did like you said to reinstall and switch to Game Ready Drivers. Sadly it did not fix the problem -_-

[MrDreadfall]

I have discovered something that could be relevant to the subject. Can someone make a sense out of this? sorry I am really a noob!

[lysxelapsed]

You can try the following: Go to your roop-unleashed\Lib\site-packages\gradio\components folder. If there's a pycache folder in it, delete it. Open up dropdown.py with an editor, change allow_custom_value: bool = False to allow_custom_value: bool = True at line 38, save the file and try again.

This has worked for a different user with that error message, but no guarantees.

[MrDreadfall]

Sadly, it didn't work. CPU is still the only dropdown option. The only difference is, I can now write in the text box, which does little because Roop still uses the CPU despite me writing "cuda" in the box.

[lysxelapsed]

Any errors when running the installer? If so, post it here. Besides that, only the usual tipps are left: Reboot your system, re-install everything needed, use the newest installer (or try the manual install explained in the wiki), have a look at the wiki, especially installation and FAQ.

[Herold-Lambda]

This should not be happening, have you installed Cudnn and the Cuda Toolkit? (ref: https://developer.nvidia.com/cuda-11-8-0-download-archive) (ref: https://developer.nvidia.com/rdp/cudnn-archive) Also, if you already updated your Nvidia drivers, have you installed it properly by doing it manually?

The step by step explanation would be:

1. Install Conda, or mini Conda
2. Git clone the repo in your desired directory
3. Open Conda and head to the directory
4. Create a Conda Environment with this code "Assuming "myenv" is not already an existing environment use this line: conda create --name myenv python=3.9 Then in the directory (path/roop-unleashed) run: pip install -r requirements.txt And then after installing everything you should be able to run it by using: python run.py

Usually cu118 torch-2.1.2 is not being installed properly installed by the windows installer, hope this helps.

[MrDreadfall]

Thank you very much ! I managed to follow every step and IT WORKS! The problem was the Window installer file all along.

[Benjamim-EP]

Isso não deveria estar acontecendo, você instalou o Cudnn e o Cuda Toolkit? (ref: https://developer.nvidia.com/cuda-11-8-0-download-archive ) (ref: https://developer.nvidia.com/rdp/cudnn-archive ) Além disso, se você já atualizou seus drivers da Nvidia, você os instalou corretamente manualmente?

A explicação passo a passo seria:

1. Instale Conda ou mini Conda
2. Git clone o repositório no diretório desejado
3. Abra o Conda e vá para o diretório
4. Crie um ambiente Conda com este código "Supondo que "myenv" ainda não seja um ambiente existente, use esta linha: conda create --name myenv python=3.9 Em seguida, no diretório (path/roop-unleashed) execute: pip install -r requirements.txt E depois de instalar tudo, você poderá executá-lo usando: python run.py

Normalmente cu118 torch-2.1.2 não está sendo instalado corretamente pelo instalador do Windows, espero que isso ajude.

God, very thanks friend