Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Jha, Animesh, Gupta, Harshit, Nandi, Ananjan
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2509.25850
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908568436867072
author Jha, Animesh
Gupta, Harshit
Nandi, Ananjan
author_facet Jha, Animesh
Gupta, Harshit
Nandi, Ananjan
contents Data selection for finetuning Large Language Models (LLMs) can be framed as a budget-constrained optimization problem: maximizing a model's downstream performance under a strict training data budget. Solving this problem is generally intractable, and existing approximate approaches are pretraining-oriented and transfer poorly to the fine-tuning setting. We reformulate this problem as a tractable Markov Decision Process (MDP) and train agents using various Reinforcement Learning (RL) methods to learn optimal data selection policies, guided by an efficient, proxy-model-based reward signal. Across four datasets, training on a $5\%$ subset selected by our approach matches or outperforms fine-tuning on the full dataset by up to $10.8$ accuracy points, while cutting wall-clock training time by up to $2 \times$, highlighting the promise of RL-guided data selection.
format Preprint
id arxiv_https___arxiv_org_abs_2509_25850
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RL-Guided Data Selection for Language Model Finetuning
Jha, Animesh
Gupta, Harshit
Nandi, Ananjan
Machine Learning
Data selection for finetuning Large Language Models (LLMs) can be framed as a budget-constrained optimization problem: maximizing a model's downstream performance under a strict training data budget. Solving this problem is generally intractable, and existing approximate approaches are pretraining-oriented and transfer poorly to the fine-tuning setting. We reformulate this problem as a tractable Markov Decision Process (MDP) and train agents using various Reinforcement Learning (RL) methods to learn optimal data selection policies, guided by an efficient, proxy-model-based reward signal. Across four datasets, training on a $5\%$ subset selected by our approach matches or outperforms fine-tuning on the full dataset by up to $10.8$ accuracy points, while cutting wall-clock training time by up to $2 \times$, highlighting the promise of RL-guided data selection.
title RL-Guided Data Selection for Language Model Finetuning
topic Machine Learning
url https://arxiv.org/abs/2509.25850