Automatically Planning Optimal Parallel Strategy for Large Language Models

TL;DR

This paper introduces an automatic parallel strategy planning algorithm for large language models that maximizes throughput by efficiently selecting optimal parallel configurations based on model and hardware data.

Contribution

It proposes a novel automatic parallel algorithm that uses a simulation model to quickly find the optimal parallel strategy for large-scale language model training.

Findings

Achieves an average of 96% accuracy in estimating training duration.

Provides globally optimal parallel strategies in real-time.

Reduces search time for parallel configurations significantly.

Abstract

The number of parameters in large-scale language models based on transformers is gradually increasing, and the scale of computing clusters is also growing. The technology of quickly mobilizing large amounts of computing resources for parallel computing is becoming increasingly important. In this paper, we propose an automatic parallel algorithm that automatically plans the parallel strategy with maximum throughput based on model and hardware information. By decoupling the training time into computation, communication, and overlap, we established a training duration simulation model. Based on this simulation model, we prune the parallel solution space to shorten the search time required. The multi-node experiment results show that the algorithm can estimate the parallel training duration in real time with an average accuracy of 96%. In our test, the recommendation strategy provided by the algorithm is always globally optimal.