TeraPipe: Token-Level Pipeline Parallelism for Training Large-Scale Language Models

TL;DR

TeraPipe introduces token-level pipeline parallelism for training large Transformer models, enabling more fine-grained and efficient training by exploiting autoregressive properties, significantly speeding up training times.

Contribution

The paper proposes a novel token-level pipeline parallelism method for Transformer models, including a dynamic programming algorithm for optimal execution scheduling.

Findings

Achieves 5.0x speedup on GPT-3 training

Enables fine-grained pipeline parallelism within sequences

Demonstrates effectiveness on large-scale models

Abstract

Model parallelism has become a necessity for training modern large-scale deep language models. In this work, we identify a new and orthogonal dimension from existing model parallel approaches: it is possible to perform pipeline parallelism within a single training sequence for Transformer-based language models thanks to its autoregressive property. This enables a more fine-grained pipeline compared with previous work. With this key idea, we design TeraPipe, a high-performance token-level pipeline parallel algorithm for synchronous model-parallel training of Transformer-based language models. We develop a novel dynamic programming-based algorithm to calculate the optimal pipelining execution scheme given a specific model and cluster configuration. We show that TeraPipe can speed up the training by 5.0x for the largest GPT-3 model with 175 billion parameters on an AWS cluster with 48 p3.16xlarge instances compared with state-of-the-art model-parallel methods. The code for reproduction can be found at https://github.com/zhuohan123/terapipe