Timezone: »

TeraPipe: Token-Level Pipeline Parallelism for Training Large-Scale Language Models
Zhuohan Li · Siyuan Zhuang · Shiyuan Guo · Danyang Zhuo · Hao Zhang · Dawn Song · Ion Stoica

Tue Jul 20 09:00 PM -- 11:00 PM (PDT) @

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

Author Information

Zhuohan Li (UC Berkeley)
Siyuan Zhuang (UC Berkeley)
Shiyuan Guo (University of California, Berkeley)
Danyang Zhuo (Duke University)
Hao Zhang (CMU)
Dawn Song (University of California, Berkeley)
Ion Stoica (UC Berkeley)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors