Glossary — Language models

Embedding

An embedding is a list of numbers representing the meaning of a piece of text, arranged so that similar meanings end up close together.

1 tokensEmbedding

Simply put#

The model turns a sentence into a row of numbers — typically a few hundred to a few thousand. Nearby rows mean nearby meaning, even when the words are completely different.

Example#

"How much does it cost?" and "What is the price?" share almost no words but land in nearly the same place as embeddings. That is what makes meaning-based search possible.

Why it matters to you#

This is the engine inside RAG. If your search has to find things by meaning rather than by exact word, embeddings are what do it.

Related terms