FineQuant: Unlocking Efficiency with Fine-Grained Weight-Only Quantization for LLMs