Bidirectional Knowledge Distillation for Enhancing Sequential Recommendation with Large Language Models