Skip to content
BENIBIX Learn

Embedding Model

Also known as Text Embedding Model, Encoder Model (Embeddings)

A model that reads text and gives back a list of numbers.

Think of it like

Think of it like a machine. It reads a note and puts a pin on the map for it.

Example

Jack's app runs 20,000 product pages through an embedding model once. Each page comes back as a list of 768 numbers, ready to search.

Why it matters

The choice of embedding model sets how well search finds the right text, and in which languages.

Definition

An embedding model is a model that turns text, or other data, into an embedding. It reads the input and gives back one list of numbers of a fixed length. It does not write text, so it is much smaller and cheaper to run than a chat model. Most are built on the encoder part of a transformer. The same model must embed both the stored texts and the questions, or the numbers will not match.
A sentence goes in and one long list of numbers comes out. The list is always the same length.

Common mistake

Moving to a new embedding model but keeping the old vectors. Each stored text has to be run through the new model again.

Key takeaways

  1. 1
    Text in, numbers out. A fixed-length list.
  2. 2
    Small and cheap. It does not write.
  3. 3
    Same model both sides. Stored text and questions.