I quantize my models with llama.cpp and it's usually one command. Some of their quants are fine-tuned by architecture but it's only to squeeze out every little performance benefit.