A1 Vertaisarvioitu alkuperäisartikkeli tieteellisessä lehdessä
Scaling Data-Constrained Language Models
Tekijät: Muennighoff, Niklas; Rush, Alexander M.; Barak, Boaz; Le Scao, Teven; Piktus, Aleksandra; Tazi, Nouamane; Pyysalo, Sampo; Wolf, Thomas; Raffel, Colin
Kustantaja: MICROTOME PUBL
Kustannuspaikka: BROOKLINE
Julkaisuvuosi: 2025
Journal: Journal of Machine Learning Research
Tietokannassa oleva lehden nimi: JOURNAL OF MACHINE LEARNING RESEARCH
Lehden akronyymi: J MACH LEARN RES
Artikkelin numero: 53
Vuosikerta: 26
Sivujen määrä: 66
ISSN: 1532-4435
eISSN: 1533-7928
Verkko-osoite: https://www.jmlr.org/papers/v26/24-1000.html
Rinnakkaistallenteen osoite: https://research.utu.fi/converis/portal/detail/Publication/492253404
The current trend of scaling language models involves increasing both parameter count and training data set size. Extrapolating this trend suggests that training data set size may soon be limited by the amount of text data available on the internet. Motivated by this limit, we investigate scaling language models in data-constrained regimes. Specifically, we run a large set of experiments varying the extent of data repetition and compute budget, ranging up to 900 billion training tokens and 9 billion parameter models. We find that with constrained data for a fixed compute budget, training with up to 4 epochs of repeated data yields negligible changes to loss compared to having unique data. However, with more repetition, the value of adding compute eventually decays to zero. We propose and empirically validate a scaling law for compute optimality that accounts for the decreasing value of repeated tokens and excess parameters. Finally, we experiment with approach esmitigating data scarcity, including augmenting the training data set with code data or removing commonly used filters. Models and data sets from our 400 training runs are freely available athttps://github.com/huggingface/datablations.
Ladattava julkaisu This is an electronic reprint of the original article. |  |
