Georg Kucsko

hmm seems to be an issue with tokenizers from hugging face. maybe just try `pip install -U tokenizers`?

added another simple option using the env var `SUNO_USE_SMALL_MODELS=True` to get smaller models that will prob fit on an 8gb card. Qw haven't implemented quantization yet. As for requirements would...

you have to set the environment variable before the model load. but also you can now more easily specify the model size in the preload function, see also here: https://github.com/suno-ai/bark/issues/51

oh yikes sorry, lemme check. feel free to also PR if you find the bug

works fine for me on a quick test, can anyone else confirm its borked?

Ah ok great ya just made some fixes there

Hm never seen that. What OS do you use? Have you tried a fresh venv?

hm, looks like precompiled encodec wheels maybe dont support your OS. maybe try installing encodec from source first

feel free to reopen if bark creates any issues

awesome, thanks sanchit. what do we need to do on the HF side for this to work?