[kkzz99]

Usually, for efficiency, you use quantized models. Quantized models reduce the number of bits available for each parameter, saving space and reduce RAM usage.