stas00/ml-engineering

Machine Learning Engineering Open Book is a free, open-source reference guide for engineers who work on training, fine-tuning, and running large AI language models (LLMs) and multimodal models (models that work with text and images together). It is written by someone who gained hands-on experience leading the training of very large open-source AI models and shares practical, copy-paste-ready solutions from those real-world projects. The book is technical in nature, it is aimed at ML engineers and infrastructure operators, not beginners to AI. It covers a wide range of practical topics organized into chapters: the hardware side (GPUs and other accelerators, memory, storage, and networking between machines in a cluster), orchestration tools for managing computing resources (including SLURM, a system widely used in research clusters), guides for actually running model training jobs, tips for running model inference efficiently, and a substantial section on debugging and troubleshooting problems that arise during large-scale training. Throughout the material you will find benchmark tools (Python scripts), comparison tables of accelerator performance and network speeds, and step-by-step guides for common tasks like testing inter-node connectivity, debugging hanging distributed training jobs, and managing cluster workloads. The book is also available as downloadable PDF and EPUB files. You would use this resource if you are responsible for training or serving large AI models on a cluster of GPUs and need practical guidance on hardware choices, network configurations, debugging, and system-level troubleshooting, rather than conceptual introductions to machine learning.