Scale MLPerf-0.6 models on Google TPU-v3 Pods

TL;DR

This paper demonstrates how to scale MLPerf-0.6 models on Google TPU-v3 Pods, achieving record performance through various optimizations and identifying challenges in model scaling.

Contribution

It introduces specific optimization techniques and insights for scaling ML models on large TPU clusters, advancing the state-of-the-art in hardware utilization.

Findings

Achieved record performance on MLPerf-0.6 benchmarks.

Identified properties that hinder model scaling.

Developed optimization strategies for large-scale TPU training.

Abstract

The recent submission of Google TPU-v3 Pods to the industry wide MLPerf v0.6 training benchmark demonstrates the scalability of a suite of industry relevant ML models. MLPerf defines a suite of models, datasets and rules to follow when benchmarking to ensure results are comparable across hardware, frameworks and companies. Using this suite of models, we discuss the optimizations and techniques including choice of optimizer, spatial partitioning and weight update sharding necessary to scale to 1024 TPU chips. Furthermore, we identify properties of models that make scaling them challenging, such as limited data parallelism and unscaled weights. These optimizations contribute to record performance in transformer, Resnet-50 and SSD in the Google MLPerf-0.6 submission.