Google DeepMind Releases Gemma 4 QAT Checkpoints

Google DeepMind released Quantization-Aware Training (QAT) checkpoints for the Gemma 4 family, targeting edge devices and consumer GPUs. The new mobile QAT schema reduces Gemma 4 E2B to about 1GB. QAT improves quality at the same size as standard PTQ, without changing memory footprint. The Q4_0 QAT format drops E2B to 3.2 GB and E4B to 5 GB.