Range-view (RV) LiDAR semantic segmentation benefits from its compatibility with established vision-based approaches, and achieves the optimal accuracy-efficiency balance for fine-grained 3D perception, which is promising for practical application. However, recently prevailing vision transformer (ViT), though pre-training on large-scale image datasets, shows weak adaption for RV representation limited by shared ViT network. To overcome above limitation to release ViT potential, we propose a novel knowledge distillation-driven vision transformer range-view LiDAR semantic segmentation framework, which aims to transfer a stronger ViT teacher prior to improve the learning capacity of compact RV-ViT student model beyond its native representation, without introducing significant additional inference cost. Firstly, considering the weak expression of compressed student feature, we design a compact context-aware RV-ViT network to dynamically integrate initial stem and earlier feature to strengthen encoding, which improves knowledge distillation learning. Then, to better transfer richer ViT teacher prior to compact RV-ViT student feature, we employ a multi-level image knowledge distillation, which comprehensively combines attention-level, feature-level and output-level similarity to encourage RV-ViT student network to focus on attention regions to promote imitation learning from ViT teacher counterpart. Finally, we conduct extensive experiments on recognized LiDAR semantic segmentation dataset SemanticKITTI and proposed method exhibits superior RV segmentation performance, which validates its ability to better utilize ViT image prior to benefit RV representation.