SuperLoRA: Parameter-Efficient Unified Adaptation of Multi-Layer Attention Modules
This work addresses the need for more flexible and efficient adaptation methods for large models in fields like NLP and computer vision, though it appears incremental as it builds upon existing LoRA variants.
The paper tackles the problem of parameter-efficient fine-tuning of large models by proposing SuperLoRA, a generalized framework that unifies and extends LoRA variants, demonstrating superior performance in transfer learning tasks, particularly in extremely few-parameter regimes.
Low-rank adaptation (LoRA) and its variants are widely employed in fine-tuning large models, including large language models for natural language processing and diffusion models for computer vision. This paper proposes a generalized framework called SuperLoRA that unifies and extends different LoRA variants, which can be realized under different hyper-parameter settings. Introducing grouping, folding, shuffling, projecting, and tensor factoring, SuperLoRA offers high flexibility compared with other LoRA variants and demonstrates superior performance for transfer learning tasks especially in the extremely few-parameter regimes.