AI Foundationspredict · compress · act

concepts → Mechanism

Mechanism

Tokenizer

decade5 connections

the rule that cuts text into tokens — it shapes everything downstream

Where it sits

The prism has six jobs across and eight layers down. Its primary cell isrepresent × L1, with a noted second cell. Hatched cells cannot exist — a GPU does not learn, an institution does not infer.

represent×L2 — tokenization is a data-layer choice, made before training

What must come first — and what it unlocks

Left to right is reading order, derived from the prerequisite_of edges. Nothing here is hand-ordered: the diagram is the graph.

nothing comes first — this is a starting pointTokenizerPretraining

Unlocks: Pretraining

What kind of thing it is — and what it is made of

Examples of it: Byte-pair encoding.

It is made of: Token, Vocabulary, Speech token.

Where to read it

The chapter that introduces it, and any chapter that uses it again.

25Breaking Language into Piecesact 6 · The Attention Revolution

Where it comes from

Every connection

All 5 edges touching this node, grouped by relation family — the sections above are highlights from this list. Colours match the relation families inthe atlas.

Structure · 4
has as a partTokenQuantity
has as an instanceByte-pair encodingAlgorithm
has as a partVocabularyDataset
has as a partSpeech tokenQuantity
Order · 1
must come beforePretrainingAlgorithm

This page is a projection of one node in src/data/concepts.ts. It has no prose file of its own — 230 declared edges produce all 236 of these pages. Edit an edge and both endpoints change.