PP-LLMs: A progressive pruning approach with medium-granularity for large language models

Kangning Du

doi:10.1016/j.eswa.2026.133770

PP-LLMs: A progressive pruning approach with medium-granularity for large language models

2026-07-31

0

PRE

AI

K

Kangning Du

Y

Yinkai Wang

B

Benkui Zhang *

J

Jinxiao Wang

F

Fan Zhang

L

Lin Cao

Y

Yanan Guo

DOI：10.1016/j.eswa.2026.133770

Original

Original request for help

Save

Abstract

En 中文

• We propose a dual-metric two-stage pipeline tailored to each layer’s function. • We introduce a method quantifying layer importance for performance-aware pruning. • We design a low-cost pruning method, PP-LLMs, for effective model deployment.

Keywords:

Large language models

Model compression

Structure pruning

Layer importance metric

PP-LLMs: A progressive pruning approach with medium-granularity for large language models

Abstract

Journal

Organization

Cited Papers

Citing Papers