← The log

drumih/turbo-fieldfare

★ 6,944 stars

SwiftLLMsmacOS

Covered in week 2026-W34 · the 2026-08 report

Gemma 4 26B-A4B inference in ~2 GB of RAM on any M-series MacBook

What turbo-fieldfare is

A custom Swift + Metal runtime for any Apple Silicon Mac, even the 8 GB ones.

From the repository README · read it on GitHub →

18 Aug 2026 · 0:24
TurboFieldfare - Run a 26B AI Model in 2GB of RAM on Any Mac
Watch on YouTube →

Transcript

TurboFieldfare squeezes a 26B model into two gigs of RAM. The Swift runtime streams only the active experts from SSD instead of loading the full 14.3 gigabyte weights. It runs on any Apple Silicon Mac with eight gigabytes as a native app or CLI, decoding between five and thirty-five tokens per second. You can run massive models on budget MacBooks.

View on GitHub