PulseAugur
EN
LIVE 09:17:25

Data repetition outperforms scaling in LLM fine-tuning, study finds

By PulseAugur Editorial · [1 sources] ·

A new research paper suggests that for supervised fine-tuning of reasoning language models, data repetition can be more effective than data scaling. The study found that training for more epochs on smaller datasets, even to the point of full memorization, yielded better generalization on benchmarks like AIME'24/25 and GPQA for the Olmo3-7B model. This approach outperformed training for a single epoch on a much larger dataset, demonstrating that token accuracy can serve as a reliable stopping criterion for SFT. AI

IMPACT Suggests a more efficient method for fine-tuning reasoning LLMs, potentially reducing data requirements and computational costs.

RANK_REASON Research paper published on arXiv detailing novel findings in LLM fine-tuning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Data repetition outperforms scaling in LLM fine-tuning, study finds

COVERAGE [1]

  1. arXiv cs.CL TIER_1 English(EN) · Dawid J. Kopiczko, Sagar Vaze, Tijmen Blankevoort, Yuki M. Asano ·

    Data Repetition Beats Data Scaling in Long-CoT Supervised Fine-Tuning

    arXiv:2602.11149v2 Announce Type: replace Abstract: Supervised fine-tuning (SFT) on chain-of-thought data is an essential post-training step for reasoning language models. Standard machine learning intuition suggests that training with more unique training samples yields better g…

RELATED ENTITIES

RELATED TOPICS