Evalute textual sentence embedding as part of a prompt

[giladgd]

I'm looking for an example of how to use both tokens and embedding as input for llama_decode.

I want to get an embedding of some sentence, and then prompt the model with the embedding and a question to make it answer based on the embedding.

Example pseudo-code:

const auto embedding = get_embedding_for("The best thing is Math");
const auto response = prompt("Fact:", embedding, "\nExplain what the fact is");

I tried looking at the way the llava example embeds an image but I couldn't replicate it with textual (tokens) embedding.
Is it possible to do?
Or is using embedding like that unique for multimodal models?