Exploring the limits of Concurrency in ML Training on Google TPUs

by   Sameer Kumar, et al.

Recent results in language understanding using neural networks have required training hardware of unprecedentedscale, with thousands of chips cooperating on a single training run. This paper presents techniques to scaleML models on the Google TPU Multipod, a mesh with 4096 TPU-v3 chips. We discuss model parallelism toovercome scaling limitations from the fixed batch size in data parallelism, communication/collective optimizations,distributed evaluation of training metrics, and host input processing scaling optimizations. These techniques aredemonstrated in both the TensorFlow and JAX programming frameworks. We also present performance resultsfrom the recent Google submission to the MLPerf-v0.7 benchmark contest, achieving record training times from16 to 28 seconds in four MLPerf models on the Google TPU-v3 Multipod machine.


Scale MLPerf-0.6 models on Google TPU-v3 Pods

The recent submission of Google TPU-v3 Pods to the industry wide MLPerf ...

ZeRO: Memory Optimization Towards Training A Trillion Parameter Models

Training large DL models with billions and potentially trillions of para...

Mesh-TensorFlow: Deep Learning for Supercomputers

Batch-splitting (data-parallelism) is the dominant distributed Deep Neur...

Parallax: Automatic Data-Parallel Training of Deep Neural Networks

The employment of high-performance servers and GPU accelerators for trai...

Training EfficientNets at Supercomputer Scale: 83 Accuracy in One Hour

EfficientNets are a family of state-of-the-art image classification mode...

Multi-node Bert-pretraining: Cost-efficient Approach

Recently, large scale Transformer-based language models such as BERT, GP...

Automatic Full Compilation of Julia Programs and ML Models to Cloud TPUs

Google's Cloud TPUs are a promising new hardware architecture for machin...