CompeteSMoE -- Effective Training of Sparse Mixture of Experts via Competition

TL;DR

This paper introduces CompeteSMoE, a novel training method for sparse mixture of experts that uses a competition-based routing mechanism to improve model scalability, efficiency, and performance across various tasks.

Contribution

The paper proposes a competition mechanism for SMoE training that addresses representation collapse and introduces CompeteSMoE, a simple, efficient algorithm with strong empirical results.

Findings

CompeteSMoE achieves superior performance over existing SMoE methods.

The competition routing policy enhances model robustness and scalability.

The proposed method maintains low computational overheads.

Abstract

Sparse mixture of experts (SMoE) offers an appealing solution to scale up the model complexity beyond the mean of increasing the network's depth or width. However, effective training of SMoE has proven to be challenging due to the representation collapse issue, which causes parameter redundancy and limited representation potentials. In this work, we propose a competition mechanism to address this fundamental challenge of representation collapse. By routing inputs only to experts with the highest neural response, we show that, under mild assumptions, competition enjoys the same convergence rate as the optimal estimator. We further propose CompeteSMoE, an effective and efficient algorithm to train large language models by deploying a simple router that predicts the competition outcomes. Consequently, CompeteSMoE enjoys strong performance gains from the competition routing policy while having low computation overheads. Our extensive empirical evaluations on two transformer architectures and a wide range of tasks demonstrate the efficacy, robustness, and scalability of CompeteSMoE compared to state-of-the-art SMoE strategies.