Fast Training of Deep Learning Models over Multiple GPUs
Citations Over TimeTop 13% of 2020 papers
Abstract
This paper proposes FastT, a transparent module to work with the TensorFlow framework for automatically identifying a satisfying deployment and execution order of operations in DNN models over multiple GPUs, for expedited model training. We propose white-box algorithms to compute the strategies with small computing resource consumption in a short time. Recently, similar studies have been done to optimize device placement using reinforcement learning. Compared to those works which learn to optimize device placement of operations in several hours using large amounts of computing resources, our approach can find excellent device placement and execution order within minutes using the same computing node as for training. We design a list of scheduling algorithms to compute the device placement and execution order for each operation and also design an algorithm to split operations in the critical path to support fine-grained (mixed) data and model parallelism to further improve the training speed in each iteration. We compare FastT with representative strategies and obtain insights on the best strategies for training different types of DNN models based on extensive testbed experiments.
Related Papers
- → Another view on parallel speedup(1990)102 cited
- → Toward a better parallel performance metric(1991)97 cited
- → Speedup for Multi-Level Parallel Computing(2012)8 cited
- → Shared virtual memory and generalized speedup(2002)12 cited
- → Ditto Subsystem : A Case Study for Building Experiment Environment in StarBED Using TopDL(2017)