Sparse High Rank Adapters
Kartikeya Bhardwaj, Nilesh Prasad Pandey, Sweta Priyadarshi, Viswanath Ganapathy, Shreya Kadambi, Rafael Esteves, Shubhankar Borse, Paul Whatmough, Risheek Garrepalli, Mart Van Baalen, Harris Teague, Markus Nagel
TL;DR
This paper introduces Sparse High Rank Adapters (SHiRA), a parameter-efficient tuning paradigm that trains highly sparse, high-rank adapters by gradient masking to modify only about 1–2% of base model weights. SHiRA enables rapid on-device adapter switching in fused mode and reduces concept loss during multi-adapter fusion, addressing key edge deployment challenges of LoRA. The authors provide both theoretical analyses of rank-sparsity tradeoffs and adapter orthogonality, and extensive experiments across vision and language tasks showing SHiRA often outperforms LoRA while enabling faster CPU loading and lower memory usage. A practical PEFT-based implementation with scatter-based weight overwriting demonstrates substantial deployment advantages, making SHiRA attractive for real-world edge and on-device applications, including SDXL/DreamBooth personalization.
Abstract
Low Rank Adaptation (LoRA) has gained massive attention in the recent generative AI research. One of the main advantages of LoRA is its ability to be fused with pretrained models, adding no overhead during inference. However, from a mobile deployment standpoint, we can either avoid inference overhead in the fused mode but lose the ability to switch adapters rapidly, or suffer significant (up to 30% higher) inference latency while enabling rapid switching in the unfused mode. LoRA also exhibits concept-loss when multiple adapters are used concurrently. In this paper, we propose Sparse High Rank Adapters (SHiRA), a new paradigm which incurs no inference overhead, enables rapid switching, and significantly reduces concept-loss. Specifically, SHiRA can be trained by directly tuning only 1-2% of the base model weights while leaving others unchanged. This results in a highly sparse adapter which can be switched directly in the fused mode. We further provide theoretical and empirical insights on how high sparsity in SHiRA can aid multi-adapter fusion by reducing concept loss. Our extensive experiments on LVMs and LLMs demonstrate that finetuning only a small fraction of the parameters in the base model significantly outperforms LoRA while enabling both rapid switching and multi-adapter fusion. Finally, we provide a latency- and memory-efficient SHiRA implementation based on Parameter-Efficient Finetuning (PEFT) Library which trains at nearly the same speed as LoRA while consuming up to 16% lower peak GPU memory, thus making SHiRA easy to adopt for practical use cases. To demonstrate rapid switching benefits during inference, we show that loading SHiRA on a base model can be 5x-16x faster than LoRA fusion on a CPU.