Pear: Pruning and Sharing Adapters in Visual Parameter-Efficient Fine-Tuning
This work addresses storage and computational inefficiencies in fine-tuning visual foundation models, offering an incremental improvement over existing adapter methods.
The paper tackles redundancy in adapters for visual parameter-efficient fine-tuning by proposing Pear, a framework that prunes less important adapters and shares important ones, achieving improved efficiency and performance on visual adaptation benchmarks.
Adapters have been widely explored to alleviate computational and storage costs when fine-tuning pretrained foundation models. However, the adapter itself can exhibit redundancy, leading to unnecessary storage overhead and inferior performance. In this paper, we propose Prune and Share (Pear), a novel adapter-pruning framework for efficient fine-tuning of pretrained visual foundation models. Specifically, we prune certain adapters and share the more important unpruned ones with positions where adapters are pruned, allowing continual adaptation at these positions after pruning. Additionally, a knowledge checkpoint strategy is introduced, which preserves the information of the pruned adapters and further boosts performance. Experimental results on visual adaptation benchmark validate the effectiveness and efficiency of the proposed Pear comparing to other competitive methods. Code is in https://github.com/yibozhong/pear.