English

Activation-Informed Pareto-Guided Low-Rank Compression for Efficient LLM/VLM

Computation and Language 2025-10-08 v1 Machine Learning

Abstract

Large language models (LLM) and vision-language models (VLM) have achieved state-of-the-art performance, but they impose significant memory and computing challenges in deployment. We present a novel low-rank compression framework to address this challenge. First, we upper bound the change of network loss via layer-wise activation-based compression errors, filling a theoretical gap in the literature. We then formulate low-rank model compression as a bi-objective optimization and prove that a single uniform tolerance yields surrogate Pareto-optimal heterogeneous ranks. Based on our theoretical insights, we propose Pareto-Guided Singular Value Decomposition (PGSVD), a zero-shot pipeline that improves activation-aware compression via Pareto-guided rank selection and alternating least-squares implementation. We apply PGSVD to both LLM and VLM, showing better accuracy at the same compression levels and inference speedup.

Keywords

Cite

@article{arxiv.2510.05544,
  title  = {Activation-Informed Pareto-Guided Low-Rank Compression for Efficient LLM/VLM},
  author = {Ryan Solgi and Parsa Madinei and Jiayi Tian and Rupak Swaminathan and Jing Liu and Nathan Susanj and Zheng Zhang},
  journal= {arXiv preprint arXiv:2510.05544},
  year   = {2025}
}
R2 v1 2026-07-01T06:20:30.778Z