---
title: "CRISP: Co-designing Pruning and Scheduling for Efficient MoE Inference on Edge Servers"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.37442v1"
permalink: "https://feed.astralor.com/items/itm_519f55cfec3e5a20353e8110"
published: "2026-10-27T15:51:14.000Z"
score: 54
channel: "tech"
tags: ["推理效率","CRISP","Co-designing Pruning","Scheduling","Efficient MoE Inference","Edge Servers Mixture-of-Experts","MoE","Although"]
---

# CRISP: Co-designing Pruning and Scheduling for Efficient MoE Inference on Edge Servers

## 速览

Mixture-of-Experts (MoE) inference on memory-constrained edge platforms is often dominated by expert-weight movement rather than computation. Although each token activates only a small subset of experts, routing is inpu…

## 关键点

- Mixture-of-Experts (MoE) inference on memory-constrained edge platforms is often dominated by expert-weight movement rather than computation.
- Although each token activates only a small subset of experts, routing is input-dependent, so all experts must remain accessible and memory movement dominates latency.
- Existing approaches such as uniform pruning and expert dropping overlook two properties of MoE execution: expert combinations are highly skewed in frequency, and experts contribute asymmetrically within each combination.

## 全文

Mixture-of-Experts (MoE) inference on memory-constrained edge platforms is often dominated by expert-weight movement rather than computation. Although each token activates only a small subset of experts, routing is input-dependent, so all experts must remain accessible and memory movement dominates latency. Existing approaches such as uniform pruning and expert dropping overlook two properties of MoE execution: expert combinations are highly skewed in frequency, and experts contribute asymmetrically within each combination. We present CRISP, a combination-aware framework that co-designs pruning and runtime scheduling for efficient MoE serving on memory-constrained edge platforms. CRISP reorders neurons by importance so one stored expert can support multiple execution widths, assigns differentiated execution to hot, warm, and cold combinations based on frequency and expert role, and exploits heterogeneous widths to overlap expert loading with computation. Across Mixtral-8x7B and DeepSeek-MoE-16B, CRISP improves throughput by up to $2.1\times$ at the same memory budget while limiting accuracy loss to 3.6\% relative to full-width inference. Our results highlight that efficient edge MoE serving requires jointly optimizing which expert capacity is preserved and how expert execution is scheduled.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_519f55cfec3e5a20353e8110