Comment by evgen

This is the way if you need speed. It costs a little bit in smarts, but compare the MTPLX option listed above with the oQ4e-mtp quant using oMLX. The good cacheing layer in oMLX will help things feel faster for some classes of tasks in my experience.