“Vision Transformers (ViTs) have made significant strides in AI-based computer vision applications. However, deploying ViTs on edge devices is challenging due to their high computational and memory demands. To tackle this, researchers have developed the TP-ViT framework, which includes a truncated uniform-log2 quantizer and a bit-decline optimization strategy. These innovations significantly reduce quantization errors and maintain model performance, even under extreme low-bit conditions. Experiments show TP-ViT outperforms state-of-the-art methods, especially in 3-bit quantization, achieving a 6.18 percentage points improvement in top-1 accuracy for ViT-small. This advancement paves the way for more efficient ViT deployment on edge hardware.”