CourseTransformers: From Attention to Trained, Served and Interpreted Models · Module 10: Efficiency and Long Context · part 71 of 111
Part 71 · Module 10: Efficiency and Long Context

Lesson 1: What is expensive, and sparse attention

27 min read·13 Oct 2026
Module 10 · Lesson 1 · About 55 minutes · Needs: Module 2, Lesson 5 (attention FLOPs, window_mask); shared blocks 1 to 9 and 15 to 19

The situation

Sol 431 at Aurora Station. Tariq Haddad, the Comms Officer, reports that long transmissions from Earth are being split. base reads 128 characters at a time, so a 4,096-character transmission becomes 32 pieces, and a question in piece 30 cannot see the answer in piece 2.

"Can it read the whole thing at once?" This lesson finds what is expensive in base, measures where attention takes over, and tests the oldest way out: let each token read only some of the others. The running example is that one transmission.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.