Kernel-level optimization for transformer models

This post is being written.