LLM-KT: Aligning Large Language Models with Knowledge Tracing using a Plug-and-Play Instruction