Mixed-Precision Embedding Using a Cache