Timezone: »

 
Poster
Mesh-TensorFlow: Deep Learning for Supercomputers
Noam Shazeer · Youlong Cheng · Niki Parmar · Dustin Tran · Ashish Vaswani · Penporn Koanantakool · Peter Hawkins · HyoukJoong Lee · Mingsheng Hong · Cliff Young · Ryan Sepassi · Blake Hechtman

Tue Dec 04 07:45 AM -- 09:45 AM (PST) @ Room 517 AB #136
in Tue Poster Session A »

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 SOTA 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

Keywords: Deep Learning · Efficient Inference Methods · Efficient Training Methods
[ Paper] 

Author Information

Noam Shazeer (Google)
Youlong Cheng (Google)

Software engineer in Tensorflow.

Niki Parmar (Google)
Dustin Tran (Google Brain)
Ashish Vaswani (Google Brain)
Penporn Koanantakool (Google)
Peter Hawkins (google.com)
HyoukJoong Lee (Google)
Mingsheng Hong (google.com)
Cliff Young (google.com)
Ryan Sepassi (Google)
Blake Hechtman (Google)

More from the Same Authors