$μ$-MoE: Test-Time Pruning as Micro-Grained Mixture-of-Experts

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Koike-Akino, Toshiaki, Liu, Jing, Wang, Ye
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915303230799872
author Koike-Akino, Toshiaki
Liu, Jing
Wang, Ye
author_facet Koike-Akino, Toshiaki
Liu, Jing
Wang, Ye
contents To tackle the huge computational demand of large foundation models, activation-aware compression techniques without retraining have been introduced. However, since these rely on calibration data, domain shift may arise for unknown downstream tasks. With a computationally efficient calibration, activation-aware pruning can be executed for every prompt adaptively, yet achieving reduced complexity at inference. We formulate it as a mixture of micro-experts, called $μ$-MoE. Several experiments demonstrate that $μ$-MoE can dynamically adapt to task/prompt-dependent structured sparsity on the fly.
format Preprint
id arxiv_https___arxiv_org_abs_2505_18451
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle $μ$-MoE: Test-Time Pruning as Micro-Grained Mixture-of-Experts
Koike-Akino, Toshiaki
Liu, Jing
Wang, Ye
Machine Learning
Artificial Intelligence
Computation and Language
To tackle the huge computational demand of large foundation models, activation-aware compression techniques without retraining have been introduced. However, since these rely on calibration data, domain shift may arise for unknown downstream tasks. With a computationally efficient calibration, activation-aware pruning can be executed for every prompt adaptively, yet achieving reduced complexity at inference. We formulate it as a mixture of micro-experts, called $μ$-MoE. Several experiments demonstrate that $μ$-MoE can dynamically adapt to task/prompt-dependent structured sparsity on the fly.
title $μ$-MoE: Test-Time Pruning as Micro-Grained Mixture-of-Experts
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.18451