Revisiting Large Language Model Pruning using Neuron Semantic Attribution

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ding, Yizhuo, Sun, Xinwei, Fu, Yanwei, Hu, Guosheng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909522439700480
author Ding, Yizhuo
Sun, Xinwei
Fu, Yanwei
Hu, Guosheng
author_facet Ding, Yizhuo
Sun, Xinwei
Fu, Yanwei
Hu, Guosheng
contents Model pruning technique is vital for accelerating large language models by reducing their size and computational requirements. However, the generalizability of existing pruning methods across diverse datasets and tasks remains unclear. Thus, we conduct extensive evaluations on 24 datasets and 4 tasks using popular pruning methods. Based on these evaluations, we find and then investigate that calibration set greatly affect the performance of pruning methods. In addition, we surprisingly find a significant performance drop of existing pruning methods in sentiment classification tasks. To understand the link between performance drop and pruned neurons, we propose Neuron Semantic Attribution, which learns to associate each neuron with specific semantics. This method first makes the unpruned neurons of LLMs explainable.
format Preprint
id arxiv_https___arxiv_org_abs_2503_01542
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Revisiting Large Language Model Pruning using Neuron Semantic Attribution
Ding, Yizhuo
Sun, Xinwei
Fu, Yanwei
Hu, Guosheng
Computation and Language
Artificial Intelligence
Model pruning technique is vital for accelerating large language models by reducing their size and computational requirements. However, the generalizability of existing pruning methods across diverse datasets and tasks remains unclear. Thus, we conduct extensive evaluations on 24 datasets and 4 tasks using popular pruning methods. Based on these evaluations, we find and then investigate that calibration set greatly affect the performance of pruning methods. In addition, we surprisingly find a significant performance drop of existing pruning methods in sentiment classification tasks. To understand the link between performance drop and pruned neurons, we propose Neuron Semantic Attribution, which learns to associate each neuron with specific semantics. This method first makes the unpruned neurons of LLMs explainable.
title Revisiting Large Language Model Pruning using Neuron Semantic Attribution
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2503.01542