not working on OSX

[davidkamp]

Hi, Would love to try this out, but I keep getting this error when following the exact install instructions on 0SX 10.15 (intel mac pro)

File "/Users/dk/opt/miniconda3/envs/audioldm/lib/python3.8/site-packages/torch/cuda/__init__.py", line 239, in _lazy_init raise AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled

I tried adding this environment variable to my .zshrc file, since that was suggested somewhere, but no difference... export PYTORCH_ENABLE_MPS_FALLBACK=1 I'd be very thankful for any help to get this running!

[microboym]

You need to run it on CPU, however, I found that in ddim.py the device is hard-coded by .cuda(). You may need to modify that line. I have tried to fix it in #13 , which is waiting for review.

[microboym]

Now you can try this

# Install ffmpeg if you haven't
brew install ffmpeg 

export PYTORCH_ENABLE_MPS_FALLBACK=1
audioldm2 -d CPU -t text 

[escoolioinglesias]

@microboym

Any difference in how you installed it vs README?

[microboym]

@escoolioinglesias You may need to pull the repo again to get the latest update from #13 and reinstall it.

cd AudioLDM2
git pull
pip install -e .

[microboym]

@escoolioinglesias Still any problem on the installation?

[escoolioinglesias]

@microboym

Everything is running great, thank you so much for your fix! <3