Exactly: you have to do all the small embeddings to get the larger embeddings.

Also, in practice, embedding are much smaller than that, the popular MiniLM v2 is 256 words.

Leaving me unsure how this helps