Reasoning Bias of Next Token Prediction Training

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Lin, Pengxiao, Zhang, Zhongwang, Xu, Zhi-Qin John
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910836581203968
author Lin, Pengxiao
Zhang, Zhongwang
Xu, Zhi-Qin John
author_facet Lin, Pengxiao
Zhang, Zhongwang
Xu, Zhi-Qin John
contents Since the inception of Large Language Models (LLMs), the quest to efficiently train them for superior reasoning capabilities has been a pivotal challenge. The dominant training paradigm for LLMs is based on next token prediction (NTP). Alternative methodologies, called Critical Token Prediction (CTP), focused exclusively on specific critical tokens (such as the answer in Q\&A dataset), aiming to reduce the overfitting of extraneous information and noise. Contrary to initial assumptions, our research reveals that despite NTP's exposure to noise during training, it surpasses CTP in reasoning ability. We attribute this counterintuitive outcome to the regularizing influence of noise on the training dynamics. Our empirical analysis shows that NTP-trained models exhibit enhanced generalization and robustness across various benchmark reasoning datasets, demonstrating greater resilience to perturbations and achieving flatter loss minima. These findings illuminate that NTP is instrumental in fostering reasoning abilities during pretraining, whereas CTP is more effective for finetuning, thereby enriching our comprehension of optimal training strategies in LLM development.
format Preprint
id arxiv_https___arxiv_org_abs_2502_02007
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Reasoning Bias of Next Token Prediction Training
Lin, Pengxiao
Zhang, Zhongwang
Xu, Zhi-Qin John
Computation and Language
Machine Learning
Since the inception of Large Language Models (LLMs), the quest to efficiently train them for superior reasoning capabilities has been a pivotal challenge. The dominant training paradigm for LLMs is based on next token prediction (NTP). Alternative methodologies, called Critical Token Prediction (CTP), focused exclusively on specific critical tokens (such as the answer in Q\&A dataset), aiming to reduce the overfitting of extraneous information and noise. Contrary to initial assumptions, our research reveals that despite NTP's exposure to noise during training, it surpasses CTP in reasoning ability. We attribute this counterintuitive outcome to the regularizing influence of noise on the training dynamics. Our empirical analysis shows that NTP-trained models exhibit enhanced generalization and robustness across various benchmark reasoning datasets, demonstrating greater resilience to perturbations and achieving flatter loss minima. These findings illuminate that NTP is instrumental in fostering reasoning abilities during pretraining, whereas CTP is more effective for finetuning, thereby enriching our comprehension of optimal training strategies in LLM development.
title Reasoning Bias of Next Token Prediction Training
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2502.02007