EntroPIC: Towards Stable Long-Term Training of LLMs via Entropy Stabilization with Proportional-Integral Control

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Yang, Kai, Xu, Xin, Chen, Yangkun, Liu, Weijie, Lyu, Jiafei, Lin, Zichuan, Ye, Deheng, Yang, Saiyong
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912865539063808
author Yang, Kai
Xu, Xin
Chen, Yangkun
Liu, Weijie
Lyu, Jiafei
Lin, Zichuan
Ye, Deheng
Yang, Saiyong
author_facet Yang, Kai
Xu, Xin
Chen, Yangkun
Liu, Weijie
Lyu, Jiafei
Lin, Zichuan
Ye, Deheng
Yang, Saiyong
contents Long-term training of large language models (LLMs) requires maintaining stable exploration to prevent the model from collapsing into sub-optimal behaviors. Entropy is crucial in this context, as it controls exploration and helps avoid premature convergence to sub-optimal solutions. However, existing reinforcement learning methods struggle to maintain an appropriate level of entropy, as the training process involves a mix of positive and negative samples, each affecting entropy in different ways across steps. To address this, we propose Entropy stabilization via Proportional-Integral Control (EntroPIC), a novel method that adaptively adjusts the influence of positive and negative samples by dynamically tuning their loss coefficients. This approach stabilizes entropy throughout training, ensuring efficient exploration and steady progress. We provide a comprehensive theoretical analysis for both on-policy and off-policy learning settings, demonstrating that EntroPIC is effective at controlling entropy in large-scale LLM training. Experimental results show that our method successfully maintains desired entropy levels, enabling stable and optimal RL training for LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2511_15248
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle EntroPIC: Towards Stable Long-Term Training of LLMs via Entropy Stabilization with Proportional-Integral Control
Yang, Kai
Xu, Xin
Chen, Yangkun
Liu, Weijie
Lyu, Jiafei
Lin, Zichuan
Ye, Deheng
Yang, Saiyong
Machine Learning
Artificial Intelligence
Long-term training of large language models (LLMs) requires maintaining stable exploration to prevent the model from collapsing into sub-optimal behaviors. Entropy is crucial in this context, as it controls exploration and helps avoid premature convergence to sub-optimal solutions. However, existing reinforcement learning methods struggle to maintain an appropriate level of entropy, as the training process involves a mix of positive and negative samples, each affecting entropy in different ways across steps. To address this, we propose Entropy stabilization via Proportional-Integral Control (EntroPIC), a novel method that adaptively adjusts the influence of positive and negative samples by dynamically tuning their loss coefficients. This approach stabilizes entropy throughout training, ensuring efficient exploration and steady progress. We provide a comprehensive theoretical analysis for both on-policy and off-policy learning settings, demonstrating that EntroPIC is effective at controlling entropy in large-scale LLM training. Experimental results show that our method successfully maintains desired entropy levels, enabling stable and optimal RL training for LLMs.
title EntroPIC: Towards Stable Long-Term Training of LLMs via Entropy Stabilization with Proportional-Integral Control
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2511.15248