Google Gemma-4-26B Ported to Apple Silicon with Half Memory Footprint

jasonkneen · x · 2026-08-24

A developer demonstrated running Google's Gemma-4-26B-A4B-it model on Apple Silicon. Using an optimized version of the MLX framework for Silicon Macs, the model loads with less than half the expected memory footprint and responds via the shell.

Original post →

More from Infra

Infra channel →