Saved in:
Bibliographic Details
Main Authors: Ru, Jinghan, Xie, Yuxin, Zhuang, Xianwei, Yin, Yuguo, Guo, Zhihui, Liu, Zhiming, Ren, Qianli, Zou, Yuexian
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.06604
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918021900009472
author Ru, Jinghan
Xie, Yuxin
Zhuang, Xianwei
Yin, Yuguo
Guo, Zhihui
Liu, Zhiming
Ren, Qianli
Zou, Yuexian
author_facet Ru, Jinghan
Xie, Yuxin
Zhuang, Xianwei
Yin, Yuguo
Guo, Zhihui
Liu, Zhiming
Ren, Qianli
Zou, Yuexian
contents Web-scale pre-training datasets are the cornerstone of LLMs' success. However, text data curated from the Internet inevitably contains random noise caused by decoding errors or unregulated web content. In contrast to previous works that focus on low quality or synthetic data, our study \textbf{provides the first systematic investigation of such random noise through a cohesive ``What-Why-How'' framework.} Surprisingly, we observed that the resulting increase in the loss of next-token prediction (NTP) was significantly lower than the proportion of random noise even when the model was scaled up to 2.7B. We provide a theoretical justification for this phenomenon, which also elucidates the success of multilingual models and can be applied to multimodal models. On the other hand, experiments show that the model's performance in downstream tasks is not based solely on the NTP loss, which means that random noise may result in degraded downstream performance. To address the potential adverse effects, we introduce a novel plug-and-play Local Gradient Matching loss, which explicitly enhances the denoising capability of the downstream task head by aligning the gradient of normal and perturbed features without requiring knowledge of the model's parameters. Additional experiments on 8 language and 14 vision benchmarks further validate its effectiveness.
format Preprint
id arxiv_https___arxiv_org_abs_2502_06604
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Do we really have to filter out random noise in pre-training data for language models?
Ru, Jinghan
Xie, Yuxin
Zhuang, Xianwei
Yin, Yuguo
Guo, Zhihui
Liu, Zhiming
Ren, Qianli
Zou, Yuexian
Computation and Language
Web-scale pre-training datasets are the cornerstone of LLMs' success. However, text data curated from the Internet inevitably contains random noise caused by decoding errors or unregulated web content. In contrast to previous works that focus on low quality or synthetic data, our study \textbf{provides the first systematic investigation of such random noise through a cohesive ``What-Why-How'' framework.} Surprisingly, we observed that the resulting increase in the loss of next-token prediction (NTP) was significantly lower than the proportion of random noise even when the model was scaled up to 2.7B. We provide a theoretical justification for this phenomenon, which also elucidates the success of multilingual models and can be applied to multimodal models. On the other hand, experiments show that the model's performance in downstream tasks is not based solely on the NTP loss, which means that random noise may result in degraded downstream performance. To address the potential adverse effects, we introduce a novel plug-and-play Local Gradient Matching loss, which explicitly enhances the denoising capability of the downstream task head by aligning the gradient of normal and perturbed features without requiring knowledge of the model's parameters. Additional experiments on 8 language and 14 vision benchmarks further validate its effectiveness.
title Do we really have to filter out random noise in pre-training data for language models?
topic Computation and Language
url https://arxiv.org/abs/2502.06604