Guardado en:
Detalles Bibliográficos
Autores principales: Wu, Haoyuan, Wang, Hai, Wu, Jiajia, Ou, Jinxiang, Wang, Keyao, Chen, Weile, Zheng, Zihao, Yu, Bei
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2512.22955
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910052297736192
author Wu, Haoyuan
Wang, Hai
Wu, Jiajia
Ou, Jinxiang
Wang, Keyao
Chen, Weile
Zheng, Zihao
Yu, Bei
author_facet Wu, Haoyuan
Wang, Hai
Wu, Jiajia
Ou, Jinxiang
Wang, Keyao
Chen, Weile
Zheng, Zihao
Yu, Bei
contents Recent advancements have shown that reinforcement learning (RL) can substantially improve the reasoning abilities of large language models (LLMs). The effectiveness of such RL training, however, depends critically on the exploration space defined by the pre-trained model's token-output distribution. In this paper, we revisit the standard cross-entropy loss, interpreting it as a specific instance of policy gradient optimization applied within a single-step episode. To systematically study how the pre-trained distribution shapes the exploration potential for subsequent RL, we propose a generalized pre-training objective that adapts on-policy RL principles to supervised learning. By framing next-token prediction as a stochastic decision process, we introduce a reward-shaping strategy that explicitly balances diversity and precision. Our method employs a positive reward scaling factor to control probability concentration on ground-truth tokens and a rank-aware mechanism that treats high-ranking and low-ranking negative tokens asymmetrically. This allows us to reshape the pre-trained token-output distribution and investigate how to provide a more favorable exploration space for RL, ultimately enhancing end-to-end reasoning performance. Contrary to the intuition that higher distribution entropy facilitates effective exploration, we find that imposing a precision-oriented prior yields a superior exploration space for RL.
format Preprint
id arxiv_https___arxiv_org_abs_2512_22955
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Diversity or Precision? A Deep Dive into Next Token Prediction
Wu, Haoyuan
Wang, Hai
Wu, Jiajia
Ou, Jinxiang
Wang, Keyao
Chen, Weile
Zheng, Zihao
Yu, Bei
Computation and Language
Recent advancements have shown that reinforcement learning (RL) can substantially improve the reasoning abilities of large language models (LLMs). The effectiveness of such RL training, however, depends critically on the exploration space defined by the pre-trained model's token-output distribution. In this paper, we revisit the standard cross-entropy loss, interpreting it as a specific instance of policy gradient optimization applied within a single-step episode. To systematically study how the pre-trained distribution shapes the exploration potential for subsequent RL, we propose a generalized pre-training objective that adapts on-policy RL principles to supervised learning. By framing next-token prediction as a stochastic decision process, we introduce a reward-shaping strategy that explicitly balances diversity and precision. Our method employs a positive reward scaling factor to control probability concentration on ground-truth tokens and a rank-aware mechanism that treats high-ranking and low-ranking negative tokens asymmetrically. This allows us to reshape the pre-trained token-output distribution and investigate how to provide a more favorable exploration space for RL, ultimately enhancing end-to-end reasoning performance. Contrary to the intuition that higher distribution entropy facilitates effective exploration, we find that imposing a precision-oriented prior yields a superior exploration space for RL.
title Diversity or Precision? A Deep Dive into Next Token Prediction
topic Computation and Language
url https://arxiv.org/abs/2512.22955