cs.LGNov 18, 2026

Intelligent Embedding Pruning for Large-Scale Recommendation Models

Authors: Anish Khazane, Alex Saad-Falcon, Guanchu Wu, Marcio Porto, Jasper Shan, Fei Yu, Eugene Shulga, Intaik Park, 

Organizations: Meta

Abstract

Modern recommendation systems face substantial memory pressure, largely from the sparse embedding layer that maps high-cardinality categorical features into dense vectors. In-training embedding pruning (ITEP) removes less important embedding rows during training, yielding end-to-end memory savings but sometimes degrading accuracy because per-table pruning ratios are set by size-based heuristics. We use feature importance (FI) to guide these ratios (FI-ITEP), assigning higher pruning to tables tied to less important features. Across three large-scale recommendation models, FI-ITEP achieves 20--40% total memory reduction while matching or improving accuracy relative to percentile-based ITEP. Because the feature-importance signal requires minimal resources to compute, these memory savings translate into concrete systems benefits: the same models train on fewer GPU trainers (e.g., 112 to 80) with higher per-GPU throughput, saving 1,000+ GPU-hours per model launch. We reproduce these trends on the open-source DLRM (Criteo 1TB) and quantify the end-to-end training-time and GPU-cost tradeoff---including the burden of computing FI---offering practical guidance for embedding pruning in real-world systems.

Figures & tables

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

CardsList
  1. Balanced Co-Clustering of Users and Items for Embedding Table Compression in Recommender Systems

    Apr 20, 2026Runhao Jiang, Renchi Yang, Donghao WuEmbedding CompressionRecommender Systems