Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac | Hacker News

TurboFieldfare, an open-source Swift and Metal inference engine for Apple Silicon Macs, can run Gemma 4 26B-A4B-IT with about 2 GB of RAM.

It keeps shared layers and the KV cache in memory while streaming routed experts from SSD on demand, making generation usable on 8 GB and 16 GB Macs.

The project also ships an experimental local OpenAI-compatible server with streaming and tool-call support.

It highlights a practical path for running large models locally on consumer Mac hardware with limited memory.