drumih/turbo-fieldfare
On-device Gemma 4 26B-A4B inference runtime for 8 GB Macs
- Chart
- Not on a chart right now
- Stars
- 7.4k
What it is
TurboFieldfare is a Swift and Metal inference runtime that runs the Gemma 4 26B-A4B model in about 2 GB of RAM on Apple Silicon Macs. It keeps the shared core and KV cache in memory while streaming only the needed experts per token from SSD.
Who it is for
Developers and Mac users who want to run a 26B-parameter model locally on 8 GB Apple Silicon machines.
Why it is trending
Version 0.10.0 was released on November 14, 2026, and the project stands out by offering a model-specific Swift and Metal runtime rather than a wrapper around MLX or llama.cpp.
Main features
- Native Mac app, CLI, and loopback OpenAI-compatible server
- Streaming model installer that repacks directly into .gturbo format
- Optional companion vision pack for image prompts on M2 or newer Macs
- Configurable expert-cache slots, prefetch, and Metal 4 runtime options
Summarized on November 17, 2026 from the README and repository data. It can be wrong; check the source before relying on it. Report a wrong summary