Comment by bogtog

When people talk about running a (quantized) medium-sized model on a Mac Mini, what types of latency and throughput times are they talking about? Do they mean like 5 tokens per second or at an actually usable speed?