What Is an Embedding Model and How Does It Work?

It is a system that converts textual information into numerical representations called vectors, which carry the meaning of the information. The more closely two pieces of text are related in meaning, the more closely their vectors sit together.

An embedding is the vector an embedding model produces for a piece of text. It is a point in a multi-dimensional space where the proximity of two points indicates similarity in meaning, which is what makes semantic search and retrieval possible.

A generation model generates text, whereas an embedding model creates a vector representation of text that already exists, for searching and comparing. In a RAG setting both collaborate: embeddings retrieve useful text, and the generation model writes the answer.

Normally with cosine similarity or a related distance measure. The smaller the distance between two vectors, the more similar the meaning of their texts, which is how a search retrieves the most relevant items.

Yes. You should embed both your documents and your queries using the same model, because their vectors then exist in the same space and the distances between them are meaningful.