GANQ: GPU-Adaptive Non-Uniform Quantization for Large Language Models | Read Paper on Bytez