Show me an embedding model that runs as fast as an optimized regex engine and I'll buy you a beer.