Fixed it for PyTorch 2.6

#253
by Fancellu - opened

https://huggingface.co/spaces/Fancellu/bark-pytorch2.6-compatible

Updated to Python 3.11, latest Gradio, updated requirements.txt, PyTorch 2.6 fix for breaking change where weights_only=True

Of course it will run v slowly on CPU, but it runs at a proper speed on a GPU

@Fancellu . This is pretty neat... although the generation times are rather slow compared to the original.

Yes. This was just a little excursion for me. Looking around I see that Kokoro is much better/faster

https://huggingface.co/spaces/Remsky/Kokoro-TTS-Zero
I also did a CPU version of this

https://huggingface.co/spaces/Fancellu/Kokoro-TTS-Zero-CPU

Faster to run on Docker

docker run -it -p 7860:7860 --platform=linux/amd64 registry.hf.space/fancellu-kokoro-tts-zero-cpu:latest python app.py

They seem to be modelled on the google voices inside aistudio

https://aistudio.google.com/generate-speech

Hi, @georg-suno ... there's an ongoing problem with this space. It queues forever. Here:
https://huggingface.co/spaces/suno/bark

jvWLXJpeNWToIeKPAJear.png

Hopefully you can put it right! Cheers x

Sign up or log in to comment