Embedding model
A model that turns a piece of text into one fixed-length vector capturing its meaning, so similarity can be measured by comparing vectors.
What people get wrong
An embedding model and a generative model are different things with different training. Embeddings from a chat model's hidden states are usually worse than those from a purpose-trained encoder a hundredth of its size.
A definition is not an understanding. This is taught properly in Turning meaning into coordinates, part of Embeddings.
Learn this properlyTerms from the same topic