HyperRouter: Towards Efficient Training and Inference of Sparse Mixture of Experts

Truong Giang Do, Le Huy Khiem, Quang Pham, TrungTin Nguyen, Thanh-Nam Doan, Binh T. Nguyen, Chenghao Liu, Savitha Ramasamy, Xiaoli Li, Steven HOI

Submission Type: Regular Short Paper

Submission Track: Efficient Methods for NLP

Submission Track 2: Theme Track: Large Language Models and the Future of NLP

Keywords: Sparse mixture of experts, hypernetwork, efficient training of LLMs, large language models

Abstract: By routing input tokens to only a few split experts, Sparse Mixture-of-Experts has enabled efficient training of large language models. Recent findings suggest that fixing the routers can achieve competitive performance by alleviating the collapsing problem, where all experts eventually learn similar representations. However, this strategy has two key limitations: (i) the policy derived from random routers might be sub-optimal, and (ii) it requires extensive resources during training and evaluation, leading to limited efficiency gains. This work introduces \texttt{HyperRouter}, which dynamically generates the router's parameters through a fixed hypernetwork and trainable embeddings to achieve a balance between training the routers and freezing them to learn an improved routing policy. Extensive experiments across a wide range of tasks demonstrate the superior performance and efficiency gains of \texttt{HyperRouter} compared to existing routing methods. Our implementation is publicly available at {\url{{https://github.com/giangdip2410/HyperRouter}}}.

Submission Number: 1522