Jhustle44/gemma-4-12B-it-qat-GGUF
Not comparable yet vs google/gemma-4-12B-it-qat-q4_0-unquantized. Not verifiable yet: the GGUF's tensors don't map onto the other model's names (3 of 677/49 match), usually an architecture llama.cpp's name map doesn't cover yet.