Modern industrial Deep Learning Recommendation Models typically extract user preferences through the analysis of sequential interaction histories, subsequently generating predictions based on these derived interests. The inherent heterogeneity in data characteristics frequently result in substantial under-utilization of computational resources during large-scale training, primarily due to computational bubbles caused by severe stragglers and slow blocking communications. This paper introduces FreeScale, a solution designed to (1) mitigate the straggler problem through meticulously load balanced input samples (2) minimize the blocking communication by overlapping prioritized embedding communications with computations (3) resolve the GPU resource competition during computation and communication overlapping by communicating through SM-Free techniques. Empirical evaluation demonstrates that FreeScale achieves up to 90.3% reduction in computational bubbles when applied to real-world workloads running on 256 H100 GPUs.
@article{arxiv.2604.24073,
title = {FreeScale: Distributed Training for Sequence Recommendation Models with Minimal Scaling Cost},
author = {Chenhao Feng and Haoli Zhang and Shakhzod Ali-Zade and Yanli Zhao and Liang Luo and Jennifer Cao and Lisen Deng and Siqiao Chen and Chenyu Zhao and Tristan Rice and Daniel Johnson and Min Si and Tiantu Xu and Yi Zhang and Siqi Yan and Chuanhao Zhuge and Min Ni and Bi Xue and Qunshu Zhang and Shen Li},
journal= {arXiv preprint arXiv:2604.24073},
year = {2026}
}
Comments
14 pages, 11 figures. Accepted to the 9th MLSys Conference, Bellevue, WA, USA, 2026