Expediting Distributed DNN Training with Device Topology-Aware Graph Deployment
This addresses the challenge of efficient DNN training in device- and topology-heterogeneous clusters, offering a practical solution for machine learning practitioners, though it is incremental as it builds on existing graph and compression methods.
The paper tackles the problem of slow distributed deep neural network (DNN) training in heterogeneous clusters by developing TAG, an automatic system that optimizes graph deployment based on device topology, achieving up to 4.56x training speed-up compared to existing schemes.
This paper presents TAG, an automatic system to derive optimized DNN training graph and its deployment onto any device topology, for expedited training in device- and topology- heterogeneous ML clusters. We novelly combine both the DNN computation graph and the device topology graph as input to a graph neural network (GNN), and join the GNN with a search-based method to quickly identify optimized distributed training strategies. To reduce communication in a heterogeneous cluster, we further explore a lossless gradient compression technique and solve a combinatorial optimization problem to automatically apply the technique for training time minimization. We evaluate TAG with various representative DNN models and device topologies, showing that it can achieve up to 4.56x training speed-up as compared to existing schemes. TAG can produce efficient deployment strategies for both unseen DNN models and unseen device topologies, without heavy fine-tuning.