openmlsys

Why we included this project

OpenMLSys is an open textbook that explains how modern machine learning systems are built from the ground up, not just how to call their APIs. Its nine chapters span the whole stack, covering tensor abstractions, automatic differentiation, GPU programming with CUDA and Triton, AI compilers, data pipelines, distributed training, model serving, and GPU cluster management. That makes it a practical reference for infrastructure engineers who need to tune or deeply customize ML stacks, and for researchers writing custom operators or scaling training across many nodes. Because it is community-maintained and freely readable online, it also works as structured self-study material for students who know ML fundamentals and want the systems perspective. If you are trying to understand how frameworks like PyTorch or MindSpore work internally, or planning your own training platform, this is a solid starting point.