05:22
2026-09-12
dev.to
ai-infrastructure
Edge0 streams MoE experts off SSD to fit 35B in 3 GB
Edge0-AI has open-sourced Edge0, an Apache 2.0 streaming inference engine that memory-maps sparse mixture-of-experts weights from SSD so a 35B model runs in roughly 2.9 GiB of peak active memory. The …