DYSL_VLA.Dysl-vla:Official code for "DySL-VLA: Efficient Vision-Language-Action Model Inference via Dynamic-Static Layer-Skipping for Robot Manipulation". A framework that dynamically skips VLA layers based on action importance, achieving 3.75× speedup and 85.7× training reduction on robotic manipulation tasks.