Compressing quantized weight

[laiviet]

Thanks for this amazing work.
I am working on how to further reduce model weight on disk.

1. Would you mind pointing me to the code (if in this repo or in other library) that saves the model weight to the disk with details about the formatting of the file? I would like to learn about how it is saved on disk.
2. is there any compression mechanism to substantially reduce the size-on-disk of the model weight? Thanks