LLM SYSTEMS ENGINEERING · LEARNING NOTES
Understand the systems
that run models.
Models, hardware and workloads.
Build understanding step by step, from shared concepts to inference, training and RL.
ModelsWhat computation and state are needed?HardwareWhere and how does execution happen?WorkloadsWhat conditions and goals matter?
Four stages, starting with small examples
01
Shared Concepts
From model computation to GPU execution
02
Inference
Serving requests efficiently with fixed weights
03
Training
Pretraining and SFT, from one GPU to multiple nodes
04
RL
Connecting generation, evaluation and learning
This site groups pretraining and SFT under training, with RL as a separate learning stage. In common terminology, SFT is also part of post-training.
Available articles
LLM Systems Engineering: Connecting Models, Hardware, and Workloads ↗
An introduction to understanding execution systems through models, hardware, and workloads, and the learning path for this series.