This AI Research Introduces Atom: A Low-Bit Quantization Technique for Efficient and Accurate Large Language Model (LLM) Serving