Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks | Read Paper on Bytez