Comment by nikolayasdf123

> non 100% correctness of kernels

wouldn't model not work properly if kernels are even slightly off?

wasn't kernels a part of training stack for models? am I missing anything?