Read-ME: Refactorizing LLMs as Router-Decoupled Mixture of Experts with System Co-Design

TL;DR

Read-ME introduces a system-aware framework to convert pre-trained dense LLMs into efficient Mixture-of-Experts models, improving inference speed and accuracy without costly retraining.

Contribution

It proposes a novel pre-gating router design that decouples expert composition from the backbone, optimizing system performance and reducing inference costs.

Findings

Achieves up to 10.1% accuracy improvement on MMLU.

Reduces mean end-to-end latency by up to 6.1%.

Enables scalable, resource-efficient LLM inference.

Abstract

The proliferation of large language models (LLMs) has led to the adoption of Mixture-of-Experts (MoE) architectures that dynamically leverage specialized subnetworks for improved efficiency and performance. Despite their benefits, MoE models face significant challenges during inference, including inefficient memory management and suboptimal batching, due to misaligned design choices between the model architecture and the system policies. Furthermore, the conventional approach of training MoEs from scratch is increasingly prohibitive in terms of cost. In this paper, we propose a novel framework Read-ME that transforms pre-trained dense LLMs into smaller MoE models (in contrast to "upcycling" generalist MoEs), avoiding the high costs of ground-up training. Our approach employs activation sparsity to extract experts. To compose experts, we examine the widely-adopted layer-wise router design and show its redundancy, and thus we introduce the pre-gating router decoupled from the MoE backbone that facilitates system-friendly pre-computing and lookahead scheduling, enhancing expert-aware batching and caching. Our codesign therefore addresses critical gaps on both the algorithmic and system fronts, establishing a scalable and efficient alternative for LLM inference in resource-constrained settings. Read-ME outperforms other popular open-source dense models of similar scales, achieving improvements of up to 10.1% on MMLU, and improving mean end-to-end latency up to 6.1%. Codes are available at: https://github.com/VITA-Group/READ-ME.