Efficient Knowledge Infusion via KG-LLM Alignment