Quantization of Google's Gemma 4 12B QAT checkpoint using a modified Q4_0 encoder that better preserves QAT weight geometry under standard GGUF FP16 block scales.
This GGUF utilizes the same encoding methodology that achieved a measured 7× reduction in KL divergence on the /Pajari/gemma-4-31B-it-qat-Q4_0_V2 model.
