Placeto: Learning Generalizable Device Placement Algorithms for Distributed Machine Learning

TL;DR

Placeto introduces a reinforcement learning method that learns generalizable device placement policies for distributed neural network training, reducing training overhead and enabling application to unseen graphs within the same family.

Contribution

It proposes a novel RL-based approach with iterative placement improvements and graph embeddings, enabling generalization across different computation graphs.

Findings

Requires up to 6.1x fewer training steps

Achieves placements on par or better than prior methods

Generalizes to unseen graphs without retraining

Abstract

We present Placeto, a reinforcement learning (RL) approach to efficiently find device placements for distributed neural network training. Unlike prior approaches that only find a device placement for a specific computation graph, Placeto can learn generalizable device placement policies that can be applied to any graph. We propose two key ideas in our approach: (1) we represent the policy as performing iterative placement improvements, rather than outputting a placement in one shot; (2) we use graph embeddings to capture relevant information about the structure of the computation graph, without relying on node labels for indexing. These ideas allow Placeto to train efficiently and generalize to unseen graphs. Our experiments show that Placeto requires up to 6.1x fewer training steps to find placements that are on par with or better than the best placements found by prior approaches. Moreover, Placeto is able to learn a generalizable placement policy for any given family of graphs, which can then be used without any retraining to predict optimized placements for unseen graphs from the same family. This eliminates the large overhead incurred by prior RL approaches whose lack of generalizability necessitates re-training from scratch every time a new graph is to be placed.