techtree / Building blocks / ← prev 12 of 26 next →
Tokens
Text is chopped into pieces, and each piece gets an ID number.
Transcript
A computer cannot read words the way you do. It only works with numbers.
So first, text is chopped into pieces. The pieces are called tokens.
Take this sentence. Here are its tokens.
Common words are usually one token each. A rare word gets split into smaller parts.
The space before a word usually sticks to that word.
There is a fixed list of every possible token, called the vocabulary. It holds tens of thousands of them.
Each token's number is just its place in that list. So the sentence becomes a list of numbers.
The same token always gets the same number. Text goes in. Numbers come out.
Needs
Nothing in the tree. This is a starting point; it assumes only arithmetic, basic algebra, functions and their graphs, exponents, square roots.
Leads to
- Embeddings video
Each token ID is looked up as a vector; similar meanings point in similar directions.