RecIS: Sparse to Dense, A Unified Training Framework for Recommendation Models
This work addresses the need for efficient and unified training frameworks for recommendation models in industrial settings, though it appears incremental as it builds on existing PyTorch optimizations.
The authors tackled the problem of training industrial-grade recommendation models by proposing RecIS, a unified sparse-dense training framework based on PyTorch, which offers superior efficiency over TensorFlow-based models and is being used in Alibaba for large-model enhanced recommendation tasks.
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.