← Tutorials5 / 22
What is an LLM?
How does a token ID become something a neural network can use?Tap to reveal
"cat" ─▶ [0.12, -0.83, 0.40, ...]
"dog" ─▶ [0.10, -0.79, 0.35, ...]
close vectors (illustrative)Through an embedding: a learned lookup table with one vector per token (thousands of numbers each).
Training moves vectors so that tokens used in similar ways end up close together. 'cat' and 'dog' end up nearer to each other than to 'carburettor'.