ParetoQ: Improving Scaling Laws in Extremely Low-bit LLM Quantization | Read Paper on Bytez