[Bug] Segmentation fault trying to generate with stable-audio-open-1.0

[NanoCode012]

Hello! Thank you for releasing the audio model.

I downloaded the model locally and ran it with

python run_gradio.py  --model-config /path/to/model_config.json --ckpt-path /path/to/model.ckpt

The gradio sets up, however, when trying to generate, it just segmentation faults.

Running on public URL: <redacted>
This share link expires in 72 hours. For free permanent hosting and GPU upgrades, run `gradio deploy` from Terminal to deploy to Spaces (https://huggingface.co/spaces)
Prompt: a hip hop song
1436392480
  0%|                                                                                                                            | 0/100 [00:00<?, ?it/s]
Segmentation fault

Env: conda with latest stable torch cuda 12.1. L4 24GB GPU.

Was anyone able to generate?

[hushaudio]

I use windows 11 with WSL2 and conda, as well as MacOS M3 Max with no issues. Generating and training both work great.

Make sure you install from the pip repo, or the git repo, but not both. That caused some issues in my environment when I first tried getting things going. Are you on google colab? Assuming not if you’re using conda.

[NanoCode012]

Thanks for your comment @hushaudio . I'll try a new env. My previous env was a bit messy as I was reusing it from sth else.

[hushaudio]

Ah yea, I did that too. That was the issue for me. Needs a clean env. I did have success adding sample diffusion to the Stable Audio env after getting generation going tho and both of them work perfectly which is great for my use case.

[NanoCode012]

Thanks! a clean env solved it.

On a clean env, I additionally had to install

# for `sndfile library not found`
sudo apt-get install libsndfile1
pip install --force-reinstall soundfile==0.10.2

# for packaging error `cannot import name 'packaging' from 'pkg_resources'`
pip install -U packaging
pip install setuptools==69.5.1

# for no flash
pip install flash-attn --no-build-isolation