Where does the embedding come from?

[xvolica]

I was very excited to find out that there is a embed function to convert a long sentence to embedding. But I didn't see the api explanation, I want to know that where does the embedding come from at llama2 model?
What more, I want to know that is there any different between using chat and embedding, what kind of prompt will generate a better embedding? Is there any issue to support that?