M-RAG: Reinforcing Large Language Model Performance through Retrieval-Augmented Generation with Multiple Partitions