Make the package installable and fix load_from_checkpoint on CPU

[gitmylo]

Did load_from_checkpoint on cpu not work? I thought using the to() method would move both the model and the weights to the correct device. Either way, better to be safe than sorry.

I see you also renamed the folder, great, I hope that projects which use it as a submodule will update then. Giving it an accurate name such as bark_hubert_quantizer instead of hubert is clearly the better name.

And adding an install with setuptools instead of requirements.txt might also be great for some users.

[rsxdalv]

Did load_from_checkpoint on cpu not work? I thought using the to() method would move both the model and the weights to the correct device. Either way, better to be safe than sorry.

I see you also renamed the folder, great, I hope that projects which use it as a submodule will update then. Giving it an accurate name such as bark_hubert_quantizer instead of hubert is clearly the better name.

And adding an install with setuptools instead of requirements.txt might also be great for some users.

That's the weird thing, I had to do both, just doing torch.load(...map_location) was not enough.