Dynamic Mixture of Experts: An Auto-Tuning Approach for Efficient Transformer Models

TL;DR

The paper introduces DynMoE, an auto-tuning method for Transformer models that dynamically adjusts expert activation, improving efficiency and performance across vision, language, and multimodal tasks.

Contribution

It proposes a novel gating and adaptive expert adjustment mechanism, enabling automatic expert selection and training efficiency in Mixture of Experts models.

Findings

Achieves competitive performance with fewer activated experts.

Reduces computational overhead during training.

Demonstrates effectiveness across diverse tasks.

Abstract

The Sparse Mixture of Experts (SMoE) has been widely employed to enhance the efficiency of training and inference for Transformer-based foundational models, yielding promising results.However, the performance of SMoE heavily depends on the choice of hyper-parameters, such as the number of experts and the number of experts to be activated (referred to as top-k), resulting in significant computational overhead due to the extensive model training by searching over various hyper-parameter configurations. As a remedy, we introduce the Dynamic Mixture of Experts (DynMoE) technique. DynMoE incorporates (1) a novel gating method that enables each token to automatically determine the number of experts to activate. (2) An adaptive process automatically adjusts the number of experts during training. Extensive numerical results across Vision, Language, and Vision-Language tasks demonstrate the effectiveness of our approach to achieve competitive performance compared to GMoE for vision and language tasks, and MoE-LLaVA for vision-language tasks, while maintaining efficiency by activating fewer parameters. Our code is available at https://github.com/LINs-lab/DynMoE.