Hey guys, i'm especially interested in feedback on the kernel design and integration onto mlx lm.

Its on pypi as well as a simple pip install mlx-turboquant :P