An embedding is the mathematical process of converting tokens (text or otherwise) into a vector of floating point numbers. This vector captures semantics such that words that have similar meaning are close to each other when their distance is measured using the metric of cosine similarity.