<SPViT: Accelerate Vision Transformer Inference on Mobile Devices via Adaptive Splitting and Offloading>