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

Jan Ludziejewski, Maciej Pióro, Jakub Krajewski, Maciej Stefaniak, Michał Krutul, Jan Małaśnicki, Marek Cygan, Piotr Sankowski, Kamil Adamczewski, Piotr Miłoś, Sebastian Jaszczur
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:41056-41073, 2025.

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. Extensive empirical validation confirms the theoretical predictions of our scaling laws. These results offer actionable insights for designing and deploying MoE models in practical large-scale training scenarios.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-ludziejewski25a, title = {Joint {M}o{E} Scaling Laws: Mixture of Experts Can Be Memory Efficient}, author = {Ludziejewski, Jan and Pi\'{o}ro, Maciej and Krajewski, Jakub and Stefaniak, Maciej and Krutul, Micha{\l} and Ma{\l}a\'{s}nicki, Jan and Cygan, Marek and Sankowski, Piotr and Adamczewski, Kamil and Mi{\l}o\'{s}, Piotr and Jaszczur, Sebastian}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {41056--41073}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/ludziejewski25a/ludziejewski25a.pdf}, url = {https://proceedings.mlr.press/v267/ludziejewski25a.html}, 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. Extensive empirical validation confirms the theoretical predictions of our scaling laws. These results offer actionable insights for designing and deploying MoE models in practical large-scale training scenarios.} }
Endnote
%0 Conference Paper %T Joint MoE Scaling Laws: Mixture of Experts Can Be Memory Efficient %A Jan Ludziejewski %A Maciej Pióro %A Jakub Krajewski %A Maciej Stefaniak %A Michał Krutul %A Jan Małaśnicki %A Marek Cygan %A Piotr Sankowski %A Kamil Adamczewski %A Piotr Miłoś %A Sebastian Jaszczur %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-ludziejewski25a %I PMLR %P 41056--41073 %U https://proceedings.mlr.press/v267/ludziejewski25a.html %V 267 %X 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. Extensive empirical validation confirms the theoretical predictions of our scaling laws. These results offer actionable insights for designing and deploying MoE models in practical large-scale training scenarios.
APA
Ludziejewski, J., Pióro, M., Krajewski, J., Stefaniak, M., Krutul, M., Małaśnicki, J., Cygan, M., Sankowski, P., Adamczewski, K., Miłoś, P. & Jaszczur, S.. (2025). Joint MoE Scaling Laws: Mixture of Experts Can Be Memory Efficient. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:41056-41073 Available from https://proceedings.mlr.press/v267/ludziejewski25a.html.

Related Material