SwapMoE: Serving Off-the-shelf MoE-based Large Language Models with Tunable Memory Budget

TL;DR

SwapMoE is a framework that enables efficient serving of large MoE-based language models on memory-limited devices by maintaining a small set of important experts, reducing memory and latency with minimal accuracy loss.

Contribution

It introduces a dynamic expert selection mechanism that allows MoE models to operate within tunable memory budgets without significant performance degradation.

Findings

Reduced memory footprint from 14.2 GiB to 4.7 GiB.

Achieved 50% latency reduction.

Maintained near-original accuracy with slight Rouge-2 score drop.

Abstract

Mixture of experts (MoE) is a popular technique to improve capacity of Large Language Models (LLMs) with conditionally-activated parallel experts. However, serving MoE models on memory-constrained devices is challenging due to the large parameter size. Typical solutions such as memory swapping or expert pruning may lead to significantly higher latency or severe accuracy loss. In this paper, we introduce SwapMoE, a framework for efficient serving of MoE-based large language models with tunable memory budgets. The main idea of SwapMoE is to keep a small dynamic set of important experts, namely Virtual Experts, in the main memory for inference, while seamlessly maintaining how the Virtual Experts map to the actual experts. Experiments have shown that SwapMoE can reduce the memory footprint while maintaining reasonable accuracy. For example, on text summarization tasks with Switch Transformer, SwapMoE can reduce the memory consumption from 14.2 GiB to 4.7 GiB, together with 50\% latency reduction and a slight Rouge-2 score drop of 0.041.