HyperMoE: Towards Better Mixture of Experts via Transferring Among Experts

TL;DR

HyperMoE introduces a hypernetwork-based MoE framework that effectively transfers knowledge among experts, balancing sparsity and expertise utilization, leading to superior performance across various datasets.

Contribution

The paper presents HyperMoE, a novel MoE model leveraging hypernetworks for knowledge transfer among experts, addressing the sparsity-performance trade-off in existing methods.

Findings

HyperMoE outperforms existing MoE methods with the same number of experts.

Knowledge transfer via HyperMoE improves model performance.

HyperMoE maintains expert selection sparsity while enhancing accuracy.

Abstract

The Mixture of Experts (MoE) for language models has been proven effective in augmenting the capacity of models by dynamically routing each input token to a specific subset of experts for processing. Despite the success, most existing methods face a challenge for balance between sparsity and the availability of expert knowledge: enhancing performance through increased use of expert knowledge often results in diminishing sparsity during expert selection. To mitigate this contradiction, we propose HyperMoE, a novel MoE framework built upon Hypernetworks. This framework integrates the computational processes of MoE with the concept of knowledge transferring in multi-task learning. Specific modules generated based on the information of unselected experts serve as supplementary information, which allows the knowledge of experts not selected to be used while maintaining selection sparsity. Our comprehensive empirical evaluations across multiple datasets and backbones establish that HyperMoE significantly outperforms existing MoE methods under identical conditions concerning the number of experts.