Joint MoE Scaling Laws: Mixture of Experts Can Be Memory Efficient

TL;DR

This paper develops joint scaling laws for mixture of experts models, revealing they can be more memory-efficient than dense models and providing a framework for optimal configuration under resource constraints.

Contribution

It introduces a theoretical framework for scaling laws of MoE models, including key factors like active parameters and dataset size, and validates predictions with extensive experiments.

Findings

MoE models can outperform dense models in memory efficiency.

Scaling laws accurately predict model performance under various constraints.

Experimental validation with over 280 tests up to 2.7B active parameters.

Abstract

Mixture of Experts (MoE) architectures have significantly increased computational efficiency in both research and real-world applications of large-scale machine learning models. However, their scalability and efficiency under memory constraints remain relatively underexplored. In this work, we present joint scaling laws for dense and MoE models, incorporating key factors such as the number of active parameters, dataset size, and the number of experts. Our findings provide a principled framework for selecting the optimal MoE configuration under fixed memory and compute budgets. Surprisingly, we show that MoE models can be more memory-efficient than dense models, contradicting conventional wisdom. To derive and validate the theoretical predictions of our scaling laws, we conduct over 280 experiments with up to 2.7B active parameters and up to 5B total parameters. These results offer actionable insights for designing and deploying MoE models in practical large-scale training scenarios.