This code is just for helping identify document topics, it literally doesn't need to be perfect. Embedding a billion documents with a server that has no GPU is neither practical nor something that yields good results.