LLM and different embeddings interaction

Maybe it’s a silly question, but I don’t understand how LLM using embeddings. There’s a multiple LLM and there’s multiple embeddings models like all-MiniLM-L6-v2 or instructor-large and some others. all of them produces different vectors (dimension) and of course content of the vector for same text. How LLM like llama2 or any other works with those different vectors ? How LLM understands what is in the vectors, as they can be produced by deferent transformers.