June 2026
Intermediate
558 pages
13h 48m
English
Building end-to-end production systems for reliable LLM serving
Everything fails all the time.
Werner Vogels, CTO of Amazon
In the previous chapters, we explored how to train large models at scale using distributed techniques like DDP, FSDP, DeepSpeed, and Megatron-LM. We also examined inference engines like vLLM and SGLang that optimize single-node and multi-node inference. Now it's time to put these pieces together into a complete production system.
AI model serving encompasses a broad spectrum of workloads. Traditional machine learning models—gradient boosted trees, linear models, and small neural networks—are typically served on CPUs with frameworks like TensorFlow ...
Read now
Unlock full access