search

gitmylo / audio-webui

A webui for different audio related Neural Networks
MIT License
1.03k stars 99 forks source link

[BUG REPORT] Unresponsive UI, Missing TensorboardX library, wrong torchvision version installed. #214

Closed d8ahazard closed 7 months ago

d8ahazard commented 7 months ago

Describe the bug So, there seem to be a handful of issues when installing on Linux. Just cloning the repo and running python ./main.py, or install.sh, or run.sh all gives the same results...

  1. After installing requirements, I get a warning that tensorboardX is not installed. This has happened for a while now, and I've just been manually installing it.

  2. I get an additional error regarding the version of TorchVision:

    raise ValueError( ValueError: Could not find the operator torchvision::nms. Please make sure you have already registered the opera tor and (if registered from C++) loaded it via torch.ops.load_library.

Reinstalling torch/torchvision/torchaudio solves this:

pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 xformers --index-url https://download.pytorch.org/whl/cu118

  1. Even after doing the above fixes, (or maybe because of them), once I load the webUI, I can't interact with anything. Selecting "suno/bark", for example, will just say "processing" forever, and nothing is shown in the logs...

To Reproduce Steps to reproduce the behavior:

  1. Install on linux manually as described in the readme...
  2. Address errors.
  3. Try to use the UI.

Expected behavior The app works.

Screenshots

image

Additional context N/A

gitmylo commented 7 months ago

Yeah, probably related to the update, I'll make sure the versions are hard set. The tensorboard message can be ignored however, it isn't used and doesn't cause any errors.

d8ahazard commented 7 months ago

@gitmylo - Sorry, I updated the original post after you responded.

I updated my version and reinstalled vanilla, no tensorboardX, no updating torch, etc. All those errors are gone.

However, the UI is still unresponsive when I try to do anything, and nothing in the logs, even with --verbose enabled.

Looking at network logs in my browser, I can see that the last message being sent is the request to load the model, then the server just...does nothing?

image
d8ahazard commented 7 months ago

GOT IT!

Updating gradio to v 3.49.0 fixed it.

Found the solution after reading #5974

Edit: I'd recommend bumping the version, and then removing the specific requirement for gradio-client, as that could potentially cause issues.

I mention this because installing 3.49.0 actually downgraded the gradio-client package to gradio_client-0.6.1, so it's likely better to just let it get installed with Gradio.

Edit2:

Will also need to remove the monkeypatch for Gradio Audio. Not sure why you were doing it, but it seems OK without it?

d8ahazard commented 7 months ago

@gitmylo This will likely fix #213 as well.