Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Yao, Yiqun, Zhang, Zheng, Li, Jing, Wang, Yequan
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2305.02869
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911828701872128
author Yao, Yiqun
Zhang, Zheng
Li, Jing
Wang, Yequan
author_facet Yao, Yiqun
Zhang, Zheng
Li, Jing
Wang, Yequan
contents Accelerating large language model pre-training is a critical issue in present research. In this paper, we focus on speeding up pre-training by progressively growing from a small Transformer structure to a large one. There are two main research problems associated with progressive growth: determining the optimal growth schedule, and designing efficient growth operators. In terms of growth schedule, the impact of each single dimension on a schedule's efficiency is under-explored by existing work. Regarding the growth operators, existing methods rely on the initialization of new weights to inherit knowledge, and achieve only non-strict function preservation, limiting further improvements on training dynamics. To address these issues, we propose Masked Structural Growth (MSG), including (i) growth schedules involving all possible dimensions and (ii) strictly function-preserving growth operators that is independent of the initialization of new weights. Experiments show that MSG is significantly faster than related work: we achieve up to 2.2x speedup in pre-training different types of language models while maintaining comparable or better downstream performances. Code is publicly available at https://github.com/cofe-ai/MSG.
format Preprint
id arxiv_https___arxiv_org_abs_2305_02869
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Masked Structural Growth for 2x Faster Language Model Pre-training
Yao, Yiqun
Zhang, Zheng
Li, Jing
Wang, Yequan
Computation and Language
Accelerating large language model pre-training is a critical issue in present research. In this paper, we focus on speeding up pre-training by progressively growing from a small Transformer structure to a large one. There are two main research problems associated with progressive growth: determining the optimal growth schedule, and designing efficient growth operators. In terms of growth schedule, the impact of each single dimension on a schedule's efficiency is under-explored by existing work. Regarding the growth operators, existing methods rely on the initialization of new weights to inherit knowledge, and achieve only non-strict function preservation, limiting further improvements on training dynamics. To address these issues, we propose Masked Structural Growth (MSG), including (i) growth schedules involving all possible dimensions and (ii) strictly function-preserving growth operators that is independent of the initialization of new weights. Experiments show that MSG is significantly faster than related work: we achieve up to 2.2x speedup in pre-training different types of language models while maintaining comparable or better downstream performances. Code is publicly available at https://github.com/cofe-ai/MSG.
title Masked Structural Growth for 2x Faster Language Model Pre-training
topic Computation and Language
url https://arxiv.org/abs/2305.02869