usamahz

Free Open-Source Book on Making ML Models Fast Released

The author released a free, open-source book, How to Make Your Model Fast, arguing that cutting FLOPs doesn't guarantee speed and that you must first identify whether a system is compute-, bandwidth-, memory- or system-bound. It covers roofline analysis, kernels, compilers, quantization, pruning, on-device LLMs, robotics, serving and agents.