WhisperX diarization fails due to PyTorch safe globals (ListConfig / TorchVersion not allowed) - M4 MacOS

[tomferrari]

Hi,

I'm using macOS with M4 chip and I got an error due to pytorch

WHISPER_DEVICE = "cpu"
WHISPER_COMPUTE_TYPE = "float32"
DIARIZATION_DEVICE = "cpu"

pyannote.audio==3.4.0
pyannote.pipeline==3.0.1
pytorch-lightning==2.6.0
torch==2.8.0
torchaudio==2.8.0
whisperx==3.7.4
python==3.13

does anyone get the same trouble ?

Thanks

WhisperX Loader échoué (Weights only load failed. This file can still be loaded, to do so you have two options, [0m[1mdo those steps only if you trust the source of the
checkpoint[0m[0m.
(1) In PyTorch 2.6, we changed the default value of the weights_only argument in torch.load from False to True. Re-running torch.load with weights_only set
to False will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source.
(2) Alternatively, to load with weights_only=True please check the recommended steps in the following error message.
WeightsUnpickler error: Unsupported global: GLOBAL omegaconf.listconfig.ListConfig was not an allowed global by default. Please use
torch.serialization.add_safe_globals() or the torch.serialization.safe_globals() context manager to allowlist this global if you trust this class/function.

Check the documentation of torch.load to learn more about types accepted by default with weights_only https://pytorch.org/docs/stable/generated/torch.load.html.).

[ShaaranSr]

I am facing the same issue in windows 11 lenovo laptop. I have i7 intel and 4050gpu but I wanted to run the whisperX on my cpu. Because my server doesnt have gpu. So building for cpu specific transcription. Even I kept the compute to float 32. I am using size "small". Still facing the same issue as you mentioned. even If I try different combinations of every version it still doesn't work. I am using python 3.11. Even I tried allowing the safe_globals still it didnt work.

Please help me if you solved your issue. Can't find solutions on youtube everyone using gpu cuda installation. Even chatgpt or any AI models whatever version it gives it still conflicts. I dont get any proper solution.

Thanks In Advance for anyone answering this!!!

[chaotic-bug]

im using linux so can go to this file
~/miniconda3/envs/whisperX/lib/python3.10/site-packages/lightning_fabric/utilities/cloud_io.py
in line 76 and change this "weights_only=weights_only" to "weights_only=False"

[TinaTe]

I tried that and it got me further, but now I get: AttributeError: module 'torchaudio' has no attribute 'AudioMetaData'
Did you encounter that as well? which version of torchaudio did you use?