Poster 146: AI Matrix: A Deep Learning Benchmark for Alibaba Data Centers

Authors: Wei Zhang (Alibaba Inc), Wei Wei (Alibaba Inc), Lingjie Xu (Alibaba Inc), Lingling Jin (Alibaba Inc), Cheng Li (University of Illinois)

Abstract: This work introduces AI Matrix, an in-house Deep Learning (DL) benchmark suite developed specifically for Alibaba's e-commerce environment. AI Matrix results from a full investigation of the DL applications used inside Alibaba and aims to cover the typical DL applications that account for more than 90% of the GPU usage in Alibaba data centers. This benchmark suite collects DL models that are either directly used or closely resemble the models used in the company’s real e-commerce applications. It also collects the real e-commerce applications if no similar DL models are not available. Through the high coverage and close resemblance to real applications, AI Matrix fully represents the DL workloads on Alibaba data centers. The collected benchmarks mainly fall into three categories: computer vision, recommendation, and language processing, which consist of the most majority of DL applications in Alibaba. AI Matrix is made open source, hoping it can benefit the public.

Best Poster Finalist (BP): no

Poster: PDF
Poster summary: PDF


Back to Poster Archive Listing