On Predictability of Reinforcement Learning Dynamics for Large Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Cai, Yuchen, Cao, Ding, Xu, Xin, Yao, Zijun, Huang, Yuqing, Tan, Zhenyu, Zhang, Benyi, Sun, Guangzhong, Liu, Guiquan, Fang, Junfeng
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914340712480768
author Cai, Yuchen
Cao, Ding
Xu, Xin
Yao, Zijun
Huang, Yuqing
Tan, Zhenyu
Zhang, Benyi
Sun, Guangzhong
Liu, Guiquan
Fang, Junfeng
author_facet Cai, Yuchen
Cao, Ding
Xu, Xin
Yao, Zijun
Huang, Yuqing
Tan, Zhenyu
Zhang, Benyi
Sun, Guangzhong
Liu, Guiquan
Fang, Junfeng
contents Recent advances in reasoning capabilities of large language models (LLMs) are largely driven by reinforcement learning (RL), yet the underlying parameter dynamics during RL training remain poorly understood. This work identifies two fundamental properties of RL-induced parameter updates in LLMs: (1) Rank-1 Dominance, where the top singular subspace of the parameter update matrix nearly fully determines reasoning improvements, recovering over 99\% of performance gains; and (2) Rank-1 Linear Dynamics, where this dominant subspace evolves linearly throughout training, enabling accurate prediction from early checkpoints. Extensive experiments across 8 LLMs and 7 algorithms validate the generalizability of these properties. More importantly, based on these findings, we propose AlphaRL, a plug-in acceleration framework that extrapolates the final parameter update using a short early training window, achieving up to 2.5 speedup while retaining \textgreater 96\% of reasoning performance without extra modules or hyperparameter tuning. This positions our finding as a versatile and practical tool for large-scale RL, opening a path toward principled, interpretable, and efficient training paradigm for LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2510_00553
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle On Predictability of Reinforcement Learning Dynamics for Large Language Models
Cai, Yuchen
Cao, Ding
Xu, Xin
Yao, Zijun
Huang, Yuqing
Tan, Zhenyu
Zhang, Benyi
Sun, Guangzhong
Liu, Guiquan
Fang, Junfeng
Machine Learning
Artificial Intelligence
Recent advances in reasoning capabilities of large language models (LLMs) are largely driven by reinforcement learning (RL), yet the underlying parameter dynamics during RL training remain poorly understood. This work identifies two fundamental properties of RL-induced parameter updates in LLMs: (1) Rank-1 Dominance, where the top singular subspace of the parameter update matrix nearly fully determines reasoning improvements, recovering over 99\% of performance gains; and (2) Rank-1 Linear Dynamics, where this dominant subspace evolves linearly throughout training, enabling accurate prediction from early checkpoints. Extensive experiments across 8 LLMs and 7 algorithms validate the generalizability of these properties. More importantly, based on these findings, we propose AlphaRL, a plug-in acceleration framework that extrapolates the final parameter update using a short early training window, achieving up to 2.5 speedup while retaining \textgreater 96\% of reasoning performance without extra modules or hyperparameter tuning. This positions our finding as a versatile and practical tool for large-scale RL, opening a path toward principled, interpretable, and efficient training paradigm for LLMs.
title On Predictability of Reinforcement Learning Dynamics for Large Language Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2510.00553