I propose adding this image to the PEFT documentation to illustrate the effectiveness of Context-Aware Prompt Tuning (CPT) in refining context embeddings for enhanced few-shot learning. The image visually supports the key concepts of CPT, such as controlled perturbations, task-specific alignment, and iterative context refinement, providing users with a clearer understanding of CPT’s advantages. Including this visual will make the documentation more comprehensive and accessible, highlighting the unique aspects of CPT in a concise and impactful way.