An embedding is the output of transforming input tokens through an embedding layer or model.
During training, an embedding layer/model will learn to map tokens to vectors in an embedding vector space, where vectors “close” to each other in this space are similar or related in some way.