---
id: 20260313-T0-18
title: "arXiv提出MoE-SpAc技术，优化异构边缘场景下的MoE推理效率"
title_en: "arXiv Introduces MoE-SpAc for Efficient Edge MoE Inference"
url: https://ai.daily.yangsir.net/daily/20260313-T0-18
issue_date: 2026-03-13
publish_date: 2026-03-12T04:00:00.000Z
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2603.09983
---

# arXiv提出MoE-SpAc技术，优化异构边缘场景下的MoE推理效率

arXiv论文arXiv:2603.09983v1提出MoE-SpAc技术，解决混合专家模型在边缘设备的内存瓶颈。通过预测专家激活的投机效用，减少I/O开销。针对自回归模型动态激活特性优化，提升边缘计算效率。

## English Version

**arXiv Introduces MoE-SpAc for Efficient Edge MoE Inference**

arXiv:2603.09983v1 proposes MoE-SpAc to address MoE memory bottlenecks on edge devices. It reduces I/O overhead by speculative utility of expert activation, optimizing for autoregressive models' dynamic nature. Improves edge computing efficiency.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2603.09983)

**详情页**：https://ai.daily.yangsir.net/daily/20260313-T0-18

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*