“The trend towards Mixture-of-Experts (MoE) models, which are larger in total size, makes running state-of-the-art models locally on consumer hardware like MacBooks infeasible.”