search

diontimmer / audio-diffusion-gradio

Decked-out gradio client for audio diffusion, mainly stable-audio-tools.
36 stars 5 forks source link

installed on windows, some issues #2

Open 3blackbar opened 1 month ago

3blackbar commented 1 month ago

I put new stable audio model with its json file to models directory - not recognized. I have only cpu as choice in "device", despite having 3090 The password thing ... not a fan. Is this train only gui? whres prompt window

rmensing commented 1 month ago

Same issues.

It requires the .ckpt version of the model, not the .safetensors. Detected mine when I put the right one in the ./models dir.

note: Once the model is detected and config file is loaded into the window on the right, you have to click on the config file and it will load the info on the left and switch you to the 'Model Process' tab.

trying now to figure out the 'device' issue.

Edit:

in gradio_extended.py lines 132-134:

        self.current_device = "cuda" if torch.cuda.is_available() else "cpu"
        self.current_device = init_device if init_device else self.current_device
        self.fp16 = True if "cuda" in self.current_device else False

just out of curiosity, since the 'Device' field is editable, I changed it to 'cuda' and got a popup error: Torch not compiled with CUDA enabled

Question is then, how do I compile Torch with CUDA enabled?

rmensing commented 1 month ago

fixed.

open cmd shell in the folder activate venv with venv\Scripts\activate

pip uninstall torch
pip cache purge
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121

close the cmd window run start_windows.bat it should now load up with 'cuda' in the device.

Gecktendo commented 1 month ago

The password isn't a requirement. You can disable that in the settings if you want.

3blackbar commented 1 month ago

Somehow torch gets borked and theres no cuda when installer runs wbui , so you need to purge and reinstall, so... nobody noticed till now that it ran on cpu ? Maybe move torchaudio install to the end of the install I cant use init audio, is ther a requirement ? Would be nice if it would save all generated sounds in a folder so we dont have to keep downloading