KronQ Makes 2-Bit LLaMA-3-70B Work Where GPTQ Degenerates

GPTQ and GPTAQ both collapse at 2-bit quantization of LLaMA-3-70B (perplexity >2,000); KronQ achieves 7.93—a result that makes extreme LLM compression practical on consumer hardware and materially changes inference deployment cost calculus.