Saved in:
Bibliographic Details
Main Authors: Yang, Qing, Peng, Qiyao, Liu, Hongtao, Liu, Kai, Qin, Bing, Liu, Ting
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.07495
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909660856975360
author Yang, Qing
Peng, Qiyao
Liu, Hongtao
Liu, Kai
Qin, Bing
Liu, Ting
author_facet Yang, Qing
Peng, Qiyao
Liu, Hongtao
Liu, Kai
Qin, Bing
Liu, Ting
contents Large Language Models (LLMs) have demonstrated exceptional performance across various tasks, with pre-training stage serving as the cornerstone of their capabilities. However, the conventional fixed-length data composition strategy for pre-training presents several practical challenges. When using shorter sequences, documents are often truncated, potentially leading to information loss and affecting the model's ability to capture long-range dependencies. Conversely, longer sequences require concatenation of multiple documents, which can introduce noise and affect the natural document boundaries and semantic coherence as well as require substantial computational overhead. To address these challenges, we first establish three quantitative metrics for evaluating data composition quality: padding ratio, truncation ratio, and concatenation ratio. Building upon these metrics, we propose a novel multi-bucket data composition method that transcends the fixed-length paradigm. Our approach adaptively organizes training data to achieve optimal composition quality as measured by the proposed metrics, offering a more flexible and efficient approach for pre-training. We conduct extensive experiments and the results demonstrate that our proposed method significantly enhances both the efficiency and effectiveness of LLM pre-training.
format Preprint
id arxiv_https___arxiv_org_abs_2407_07495
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Beyond Fixed Length: Bucket Pre-training is All You Need
Yang, Qing
Peng, Qiyao
Liu, Hongtao
Liu, Kai
Qin, Bing
Liu, Ting
Computation and Language
Large Language Models (LLMs) have demonstrated exceptional performance across various tasks, with pre-training stage serving as the cornerstone of their capabilities. However, the conventional fixed-length data composition strategy for pre-training presents several practical challenges. When using shorter sequences, documents are often truncated, potentially leading to information loss and affecting the model's ability to capture long-range dependencies. Conversely, longer sequences require concatenation of multiple documents, which can introduce noise and affect the natural document boundaries and semantic coherence as well as require substantial computational overhead. To address these challenges, we first establish three quantitative metrics for evaluating data composition quality: padding ratio, truncation ratio, and concatenation ratio. Building upon these metrics, we propose a novel multi-bucket data composition method that transcends the fixed-length paradigm. Our approach adaptively organizes training data to achieve optimal composition quality as measured by the proposed metrics, offering a more flexible and efficient approach for pre-training. We conduct extensive experiments and the results demonstrate that our proposed method significantly enhances both the efficiency and effectiveness of LLM pre-training.
title Beyond Fixed Length: Bucket Pre-training is All You Need
topic Computation and Language
url https://arxiv.org/abs/2407.07495