Accuracy-Performance Trade-Offs in LLM Quantization