Comment by trebligdivad

Even local, MoE are just so much faster, and they let you pick a large/less quantized model and still get a useful speed.