Theano-MPI: a Theano-based Distributed Training Framework
This work addresses the need for efficient distributed training in deep learning, particularly for researchers and practitioners using Theano, but it is incremental as it builds on existing data parallelism and MPI concepts.
The authors tackled the problem of scaling deep learning training across multiple GPUs by developing Theano-MPI, a distributed training framework that supports both synchronous and asynchronous training using CUDA-aware MPI for parameter exchange. They demonstrated the framework's ability to reduce training time when scaling AlexNet and GoogLeNet from 2 to 8 GPUs, and released it as open-source.
We develop a scalable and extendable training framework that can utilize GPUs across nodes in a cluster and accelerate the training of deep learning models based on data parallelism. Both synchronous and asynchronous training are implemented in our framework, where parameter exchange among GPUs is based on CUDA-aware MPI. In this report, we analyze the convergence and capability of the framework to reduce training time when scaling the synchronous training of AlexNet and GoogLeNet from 2 GPUs to 8 GPUs. In addition, we explore novel ways to reduce the communication overhead caused by exchanging parameters. Finally, we release the framework as open-source for further research on distributed deep learning