1 paper
Simo Ryu, Seunghyun Seo, Jaejun Yoo
In this paper, we present an efficient method for storing fine-tuned models by leveraging the low-rank properties of weight residuals. Our key observation is that weight residuals…