No numbers. Not sure if anybody cares… I’ve run the UD version of Q4_k_m for a month. I talk to this model nicely, because it’s a functional nervous wreck. And initially I thought that might be an alignment thing, so I also have the heretic version when I need a breather from this hyper vigilant over achiever llm. Don’t get me wrong. It’s great ! Works well .. most of the time. It’s when the context gets long(in my case 20k!), chain of tools gets long , or it knows that it previous has made...