Tutorial Slides
Beamer decks for the MLSys·im conference tutorial on analytical modeling of ML systems.
The MLSys·im tutorial is delivered as a Beamer deck in four modules. The PDFs are built from source on every release (see mlsysim/tutorial/slides/) and published alongside this site.
Module 1: Foundations & Architecture
Foundations: the iron law of ML systems, the roofline model, constraint-based systems engineering, dimensional strictness, and the Pydantic registry system.
Module 2: Advanced Single-Node Analysis
Bottlenecks: the data wall (ingestion vs transformation), LLM serving limits (pre-fill vs decoding), algorithmic optimizations (speculative decoding), and the reasoning wall (OpenAI o1 scaling).
Module 3: Scale, Dollars, and Carbon
Fleet dynamics: distributed training, the Amdahl Trap, Total Cost of Ownership (the infrastructure multiplier), and cross-domain carbon accounting.
Module 4: Design Space Exploration & Synthesis
Systems engineering: rapid parametric sweeps, the universality of the roofline (TinyML to Frontier), sensitivity analysis, and SLA-driven hardware synthesis.
Source
The decks live in the mlsysim/tutorial/slides/ directory of the MLSysBook repository. To rebuild locally:
cd mlsysim/tutorial
makeThis produces the four PDFs using pdflatex and the shared beamerthememlsys theme. Prerequisites: pdflatex (TeX Live or MacTeX) plus either inkscape or rsvg-convert for SVG conversion.
Textbook lecture slides
The 35-deck textbook lecture series covering both volumes of Machine Learning Systems lives at mlsysbook.ai/slides. Those decks are organized by chapter and are independent of the MLSys·im tutorial.