---
id: 20260812-T0-05
title: "LorExperts与BTExperts：MoE专家压缩新方法，保留精度同时降低成本"
title_en: "LorExperts and BTExperts: Shape-Mutating Compression for MoE Models"
url: https://ai.daily.yangsir.net/daily/20260812-T0-05
issue_date: 2026-08-12
publish_date: 2026-08-11T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2608.07814
---

# LorExperts与BTExperts：MoE专家压缩新方法，保留精度同时降低成本

arXiv新论文提出LorExperts和BTExperts两种MoE专家压缩方法。现有专家剪枝或合并方法（如REAP）会在降低成本时牺牲精度。新方法通过形状变异（shape mutating）技术，在压缩专家权重矩阵的同时保持模型准确性，为大规模MoE模型的低成本部署提供新方案。

## English Version

**LorExperts and BTExperts: Shape-Mutating Compression for MoE Models**

New paper introduces LorExperts and BTExperts, shape-mutating compression methods for MoE expert weight matrices. Unlike existing pruning/merging approaches that sacrifice accuracy, these methods maintain model quality while enabling cost-efficient deployment of large MoE models.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2608.07814)

**详情页**：https://ai.daily.yangsir.net/daily/20260812-T0-05

---

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