[DrJosiah]

It might have been a typo, as the current llama.cpp / alpaca.cpp included quantizers default to 4 bits.