Mesh-TensorFlow: Deep Learning for Supercomputers

TL;DR

Mesh-TensorFlow introduces a flexible language for specifying distributed tensor computations, enabling efficient training of large models across supercomputers, surpassing previous state-of-the-art results in translation and language modeling.

Contribution

It presents Mesh-TensorFlow, a novel language for defining general tensor distribution strategies, facilitating scalable model and data parallelism on large clusters.

Findings

Trained Transformer models with up to 5 billion parameters.

Achieved state-of-the-art results on WMT'14 translation.

Demonstrated efficient training on TPU meshes of up to 512 cores.

Abstract

Batch-splitting (data-parallelism) is the dominant distributed Deep Neural Network (DNN) training strategy, due to its universal applicability and its amenability to Single-Program-Multiple-Data (SPMD) programming. However, batch-splitting suffers from problems including the inability to train very large models (due to memory constraints), high latency, and inefficiency at small batch sizes. All of these can be solved by more general distribution strategies (model-parallelism). Unfortunately, efficient model-parallel algorithms tend to be complicated to discover, describe, and to implement, particularly on large clusters. We introduce Mesh-TensorFlow, a language for specifying a general class of distributed tensor computations. Where data-parallelism can be viewed as splitting tensors and operations along the "batch" dimension, in Mesh-TensorFlow, the user can specify any tensor-dimensions to be split across any dimensions of a multi-dimensional mesh of processors. A Mesh-TensorFlow graph compiles into a SPMD program consisting of parallel operations coupled with collective communication primitives such as Allreduce. We use Mesh-TensorFlow to implement an efficient data-parallel, model-parallel version of the Transformer sequence-to-sequence model. Using TPU meshes of up to 512 cores, we train Transformer models with up to 5 billion parameters, surpassing state of the art results on WMT'14 English-to-French translation task and the one-billion-word language modeling benchmark. Mesh-Tensorflow is available at https://github.com/tensorflow/mesh .