Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac
Summary
The article discusses TurboFieldfare, an open-source runtime that enables Gemma 4 26B-A4B inference on Apple Silicon Macs using around 2 GB of RAM by streaming weights from SSD. It provides a native Mac app, CLI, and a local OpenAI-compatible server, all implemented in Swift/Metal, with documented benchmarks and memory-management approaches.