About This Resource

An experimental inference tool for running mixture-of-experts language models on Macs by streaming expert weights from SSD storage. It explores trading storage access for a lower memory requirement.