Saved in:
Bibliographic Details
Main Authors: Ren, Jie, Ma, Bin, Yang, Shuangyan, Francis, Benjamin, Ardestani, Ehsan K., Si, Min, Li, Dong
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.08568
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908644971380736
author Ren, Jie
Ma, Bin
Yang, Shuangyan
Francis, Benjamin
Ardestani, Ehsan K.
Si, Min
Li, Dong
author_facet Ren, Jie
Ma, Bin
Yang, Shuangyan
Francis, Benjamin
Ardestani, Ehsan K.
Si, Min
Li, Dong
contents Deep learning recommendation models (DLRMs) are widely used in industry, and their memory capacity requirements reach the terabyte scale. Tiered memory architectures provide a cost-effective solution but introduce challenges in embedding-vector placement due to complex embedding-access patterns. We propose RecMG, a machine learning (ML)-guided system for vector caching and prefetching on tiered memory. RecMG accurately predicts accesses to embedding vectors with long reuse distances or few reuses. The design of RecMG focuses on making ML feasible in the context of DLRM inference by addressing unique challenges in data labeling and navigating the search space for embedding-vector placement. By employing separate ML models for caching and prefetching, plus a novel differentiable loss function, RecMG narrows the prefetching search space and minimizes on-demand fetches. Compared to state-of-the-art temporal, spatial, and ML-based prefetchers, RecMG reduces on-demand fetches by 2.2x, 2.8x, and 1.5x, respectively. In industrial-scale DLRM inference scenarios, RecMG effectively reduces end-to-end DLRM inference time by up to 43%.
format Preprint
id arxiv_https___arxiv_org_abs_2511_08568
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Machine Learning-Guided Memory Optimization for DLRM Inference on Tiered Memory
Ren, Jie
Ma, Bin
Yang, Shuangyan
Francis, Benjamin
Ardestani, Ehsan K.
Si, Min
Li, Dong
Performance
Deep learning recommendation models (DLRMs) are widely used in industry, and their memory capacity requirements reach the terabyte scale. Tiered memory architectures provide a cost-effective solution but introduce challenges in embedding-vector placement due to complex embedding-access patterns. We propose RecMG, a machine learning (ML)-guided system for vector caching and prefetching on tiered memory. RecMG accurately predicts accesses to embedding vectors with long reuse distances or few reuses. The design of RecMG focuses on making ML feasible in the context of DLRM inference by addressing unique challenges in data labeling and navigating the search space for embedding-vector placement. By employing separate ML models for caching and prefetching, plus a novel differentiable loss function, RecMG narrows the prefetching search space and minimizes on-demand fetches. Compared to state-of-the-art temporal, spatial, and ML-based prefetchers, RecMG reduces on-demand fetches by 2.2x, 2.8x, and 1.5x, respectively. In industrial-scale DLRM inference scenarios, RecMG effectively reduces end-to-end DLRM inference time by up to 43%.
title Machine Learning-Guided Memory Optimization for DLRM Inference on Tiered Memory
topic Performance
url https://arxiv.org/abs/2511.08568