github.com
TurboFieldfare
Gemma 4 26B inference in ~2 GB RAM on Apple Silicon.
About TurboFieldfare
TurboFieldfare provides a Swift + Metal runtime to run Gemma 4 26B inference on Apple Silicon Macs with about 2 GB RAM. It streams only the necessary experts from SSD instead of loading the full 14.3 GB model into memory. The project includes a runtime, streaming installer, CLI, and native Mac app written in Swift and Metal.
Categories
Topics
Published
Discussion highlights
AI-generated summary based on Hacker News comments
🔗 Full discussion: https://news.ycombinator.com/item?id=49098510
No comments yet. Have you tried it?