paper-with-me

Papers

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

2025-05-24 · Toshiaki Koike-Akino, Jing Liu, Ye Wang

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 $\mu$-MoE. Several experiments demonstrate that $\mu$-MoE can dynamically adapt to task/prompt-dependent structured sparsity on the fly.

📄 PDF Abstract BibTeX arXiv:2505.18451

Code (0)

등록된 구현이 없습니다.

Tasks

Mixture-of-Experts

Methods 이 논문이 사용한 방법론

Pruning 설명 없음

Similar Papers 제목 키워드 기반

CAMERA: Multi-Matrix Joint Compression for MoE Models via Micro-Expert Redundancy Analysis

2025-08-04 · Yuzhuang Xu, Xu Han, Yuanchi Zhang, Yixuan Wang 외 arxiv

Large Language Models (LLMs) with Mixture-of-Experts (MoE) architectures are distinguished by their strong performance scaling with increasing parameters across a wide range of tasks, yet they also suffer from substantia…

Computational Efficiency

UnIT: Scalable Unstructured Inference-Time Pruning for MAC-efficient Neural Inference on MCUs

2025-07-10 · Ashe Neth, Sawinder kaur, Mohammad Nur Hossain Khan, Subrata Biswas 외 arxiv

Existing pruning methods are typically applied during training or compile time and often rely on structured sparsity. While compatible with low-power microcontrollers (MCUs), structured pruning underutilizes the opportun…

Attribution-Guided and Coverage-Maximized Pruning for Structural MoE Compression

2026-06-16 · Yifu Ding, Jiacheng Wang, Ge Yang, Yongcheng Jing 외 arxiv

Mixture-of-Experts (MoE) models scale compute efficiently, yet remain expensive to deploy due to their substantial memory footprint and inference overhead. Prior compression methods mainly operate at the expert level, ei…

POP: Online Structural Pruning Enables Efficient Inference of Large Foundation Models

2026-02-06 · Yi Chen, Wonjin Shin, Shuhong Liu, Tho Mai 외 arxiv

Large foundation models (LFMs) achieve strong performance through scaling, yet current structural pruning methods derive fixed pruning decisions during inference, overlooking sparsity patterns that emerge in the autoregr…

MiCRo: Mixture Modeling and Context-aware Routing for Personalized Preference Learning

2025-05-30 · Jingyan Shen, Jiarui Yao, Rui Yang, Yifan Sun 외

Reward modeling is a key step in building safe foundation models when applying reinforcement learning from human feedback (RLHF) to align Large Language Models (LLMs). However, reward modeling based on the Bradley-Terry …