English

Learning Scalable Model Soup on a Single GPU: An Efficient Subspace Training Strategy

Machine Learning 2024-07-24 v2

Abstract

Pre-training followed by fine-tuning is widely adopted among practitioners. The performance can be improved by "model soups"~\cite{wortsman2022model} via exploring various hyperparameter configurations.The Learned-Soup, a variant of model soups, significantly improves the performance but suffers from substantial memory and time costs due to the requirements of (i) having to load all fine-tuned models simultaneously, and (ii) a large computational graph encompassing all fine-tuned models. In this paper, we propose Memory Efficient Hyperplane Learned Soup (MEHL-Soup) to tackle this issue by formulating the learned soup as a hyperplane optimization problem and introducing block coordinate gradient descent to learn the mixing coefficients. At each iteration, MEHL-Soup only needs to load a few fine-tuned models and build a computational graph with one combined model. We further extend MEHL-Soup to MEHL-Soup+ in a layer-wise manner. Experimental results on various ViT models and data sets show that MEHL-Soup(+) outperforms Learned-Soup(+) in terms of test accuracy, and also reduces memory usage by more than 13×13\times. Moreover, MEHL-Soup(+) can be run on a single GPU and achieves 9×9\times speed up in soup construction compared with the Learned-Soup. The code is released at https://github.com/nblt/MEHL-Soup.

Keywords

Cite

@article{arxiv.2407.03641,
  title  = {Learning Scalable Model Soup on a Single GPU: An Efficient Subspace Training Strategy},
  author = {Tao Li and Weisen Jiang and Fanghui Liu and Xiaolin Huang and James T. Kwok},
  journal= {arXiv preprint arXiv:2407.03641},
  year   = {2024}
}

Comments

ECCV 2024

R2 v1 2026-06-28T17:28:46.365Z