Google Gemma 4 Models Optimized for Efficiency

Google has released new versions of its Gemma 4 family of AI models, optimized with Quantization-Aware Training (QAT) to significantly reduce memory requirements and enhance on-device performance. This advancement allows the models to run efficiently on everyday edge devices and consumer GPUs, minimizing quality loss through integrated quantization during training. The new checkpoints, including a novel mobile-specific format, drastically cut memory footprints, making Gemma 4 E2B models require less than 1GB of memory, thereby improving accessibility and usability for developers across various platforms.