I’ve been using gpt-oss-120B with CPU MoE offloading on a 24GB GPU and it’s very usable. Excited to see if I can get good results on this now!