\( \newcommand{\vect}[1]{\mathbf{#1}} \newcommand{\mat}[1]{\mathbf{#1}} \newcommand{\R}{\mathbb{R}} \newcommand{\E}{\mathbb{E}} \newcommand{\Prob}{\mathbb{P}} \newcommand{\T}{^{\top}} \newcommand{\norm}[1]{\lVert #1 \rVert} \newcommand{\abs}[1]{\lvert #1 \rvert} \newcommand{\ihat}{\mkern1mu\hat{\boldsymbol{\imath}}\mkern2mu} \newcommand{\jhat}{\mkern1mu\hat{\boldsymbol{\jmath}}\mkern2mu} \newcommand{\pd}[2]{\frac{\partial #1}{\partial #2}} \newcommand{\argmax}{\operatorname*{arg\,max}} \newcommand{\argmin}{\operatorname*{arg\,min}} \newcommand{\cdata}[1]{\class{c-data}{#1}} \newcommand{\cmodel}[1]{\class{c-model}{#1}} \newcommand{\ctruth}[1]{\class{c-truth}{#1}} \newcommand{\cgrad}[1]{\class{c-grad}{#1}} \newcommand{\closs}[1]{\class{c-loss}{#1}} \newcommand{\cparam}[1]{\class{c-param}{#1}} \newcommand{\csecond}[1]{\class{c-second}{#1}} \renewcommand{\,}{\mkern3mu} \)

E04 · Large-Scale Training Systems

A free, illustrated L4 course of 8 modules in the ML Engineering track. Modules include: Data, model and pipeline parallelism; Distributed data loading; Gradient synchronization and communication.

E04 L4 · Expert 8 modules · about 32 lessons Planned · Ring 4

Part of the ML Engineering track. Ships in Ring 4 (Expert) of the roadmap.

Modules

1. Data, model and pipeline parallelism
2. Distributed data loading
3. Gradient synchronization and communication
4. Checkpointing and fault tolerance
5. Memory: activation checkpointing and sharding
6. Cluster scheduling and resource accounting
7. Training LLMs at scale
8. Profiling and finding bottlenecks

Signature illustrations

The pictures a learner should still see in their head a year later:

Parallelism topologies, memory budget bars, throughput scaling curves.

Status

This course is planned, not written yet. It ships with Ring 4 (Expert).

Every lesson will follow the same twelve-part blueprint with at least three purpose-made figures. Watch the repository to hear when modules ship, or help write and review it.