In this paper, we propose RecIS, a unified Sparse-Dense training framework designed to achieve two primary goals: 1. Unified Framework To create a Unified sparse-dense training framework based on the PyTorch ecosystem that meets the training needs of industrial-grade recommendation models that integrated with large models. 2.System Optimization To optimize the sparse component, offering superior efficiency over the TensorFlow-based recommendation models. The dense component, meanwhile, leverages existing optimization technologies within the PyTorch ecosystem. Currently, RecIS is being used in Alibaba for numerous large-model enhanced recommendation training tasks, and some traditional sparse models have also begun training in it.
@article{arxiv.2509.20883,
title = {RecIS: Sparse to Dense, A Unified Training Framework for Recommendation Models},
author = {Hua Zong and Qingtao Zeng and Zhengxiong Zhou and Zhihua Han and Zhensong Yan and Mingjie Liu and Hechen Sun and Jiawei Liu and Yiwen Hu and Qi Wang and YiHan Xian and Wenjie Guo and Houyuan Xiang and Zhiyuan Zeng and Xiangrong Sheng and Bencheng Yan and Nan Hu and Yuheng Huang and Jinqing Lian and Ziru Xu and Yan Zhang and Ju Huang and Siran Yang and Huimin Yi and Jiamang Wang and Pengjie Wang and Han Zhu and Jian Wu and Dan Ou and Jian Xu and Haihong Tang and Yuning Jiang and Bo Zheng and Lin Qu},
journal= {arXiv preprint arXiv:2509.20883},
year = {2025}
}