SuperLoRA: Parameter-Efficient Unified Adaptation of Multi-Layer Attention Modules
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913269957001216 |
|---|---|
| author | Chen, Xiangyu Liu, Jing Wang, Ye Wang, Pu Perry Brand, Matthew Wang, Guanghui Koike-Akino, Toshiaki |
| author_facet | Chen, Xiangyu Liu, Jing Wang, Ye Wang, Pu Perry Brand, Matthew Wang, Guanghui Koike-Akino, Toshiaki |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2403_11887 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | SuperLoRA: Parameter-Efficient Unified Adaptation of Multi-Layer Attention Modules Chen, Xiangyu Liu, Jing Wang, Ye Wang, Pu Perry Brand, Matthew Wang, Guanghui Koike-Akino, Toshiaki Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning 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. |
| title | SuperLoRA: Parameter-Efficient Unified Adaptation of Multi-Layer Attention Modules |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2403.11887 |