Edge0 Enables 35B AI Model to Run on Mac mini with Under 3GB Memory Using SSD Streaming

Researchers have released Edge0, an open-source inference framework that runs a 35-billion-parameter language model on a Mac mini M4 Pro using just 2.9 GB of active memory—a 90% reduction compared to loading the full model into RAM. The approach achieves this by streaming model weights from an SSD on demand, combined with a trained routing predictor that fetches experts one step ahead of computation.