Merging Experts into One: Improving Computational Efficiency of Mixture of Experts
Submission Type: Regular Short Paper
Submission Track: Efficient Methods for NLP
Keywords: Mixture of Experts, Computational Efficiency
TL;DR: In this work, we propose merging experts into one (MEO) as a drop-in replacement of the mixture of experts (MoE) to significantly improve the computational efficiency, e.g., reducing FLOPs from 72.0G to 28.6G.
Abstract: Scaling the size of language models usually leads to remarkable advancements in NLP tasks. But it often comes with a price of growing computational cost. Although a sparse Mixture of Experts (MoE) can reduce the cost by activating a small subset of parameters (e.g., one expert) for each input, its computation escalates significantly if increasing the number of activated experts, limiting its practical utility. Can we retain the advantages of adding more experts without substantially increasing the computational costs? In this paper, we first demonstrate the superiority of selecting multiple experts and then propose a computation-efficient approach called \textbf{\texttt{Merging Experts into One}} (MEO), which reduces the computation cost to that of a single expert. Extensive experiments show that MEO significantly improves computational efficiency, e.g., FLOPS drops from 72.0G of vanilla MoE to 28.6G (MEO). Moreover, we propose a token-level attention block that further enhances the efficiency and performance of token-level MEO, e.g., 83.3\% (MEO) vs. 82.6\% (vanilla MoE) average score on the GLUE benchmark. Our code will be released upon acceptance. Code will be released at: \url{https://github.com/Shwai-He/MEO}.
Submission Number: 1905
Loading