A native Mac app, CLI, and local server for a 26B model on 8 GB Apple Silicon is more interesting because the constraints are explicit. The README says where the model lives, when images need a separate pack, and that only one model-owning app should run at once.
TurboFieldfare
Efficient local inference of Gemma 4 26B-A4B model on Apple Silicon Macs with ~2 GB RAM
80 ratings4.27
Saves919
CategoryAI
PlatformWeb
PricingFree
TurboFieldfare is a custom Swift and Metal runtime that enables inference of the instruction-tuned Gemma 4 26B-A4B large language model on Apple Silicon Macs using approximately 2 GB of RAM. It streams model expert weights from SSD as needed, allowing operation even on Macs with 8 GB RAM. It includes a native Mac app, CLI, and a local OpenAI-compatible server for chat completions with support for images via a companion vision pack.
Ratings & Reviews
Write a reviewInformation
CategoryAI
PlatformsWeb
PricingFree
AccountNot confirmed
ListedJuly 2026
WebVisit
Report a problem
Around the Web
Show HN919 source points · 0 comments