Google’s Gemma 4 Models Just Got a Memory Diet — Without the Usual Quality Crash
Google’s new Gemma 4 model checkpoints use quantization-aware training to slash memory usage while maintaining quality — making on-device AI more practical across phones and laptops.