Tag
ShadowPEFT introduces a centralized parameter-efficient fine-tuning method that uses a depth-shared shadow module to refine transformer layer representations, matching or outperforming LoRA/DoRA with comparable trainable parameters.