What parameter sizes are you talking about? The 30B-A3B is very slow on my device even at low quantizations. The 4B version works, but isn't very good, and the 80B-A3B doesn't even fit!