Module 1 of 11
LLM foundations
- Beginner
- 3 lessons
- 1 h 25 min
Chapter 1 · The Lowtide Sleeper
Dusk at Harbour Terminus, and the Engine Room is warming up. Tobias, the announcer the crew calls the Voice, makes his first announcement and it is clear and correct, so everyone relaxes. A little later he names a station that is not on the line, in exactly the same voice. In this chapter you will see how a sentence is made one piece at a time, and why it can be wrong without anything being broken.
What you will be able to do
- Explain how an LLM generates text: attention, the KV-cache, tokenization and sampling.
- Choose sensible sampling settings and compare open and closed models.
Lessons
How LLMs workCompleted
Tokenization and embeddingsCompleted
Read first: How LLMs work
Sampling parameters and the model landscapeCompleted
Read first: How LLMs work
End of chapter 1 · The Lowtide Sleeper
The board found room for every name, Tobias read them out, and the Lowtide Sleeper left Harbour Terminus on time, more or less.