Fine-tuning pre-trained vision models for downstream tasks is cornerstone of computer vision, enabling the adaptation of general-purpose models to specialized applications. However, the growing size ...