I ran GLM-5.1 on a 16GB RAM machine

Beautiful. Twenty seconds per token is funny, but the only direction is up from here!