@inproceedings{ulm-etal-2025-contrastive,
title = "Contrastive Decoding for Synthetic Data Generation in Low-Resource Language Modeling",
author = "Ulm, Jannek and
Du, Kevin and
Sn{\ae}bjarnarson, V{\'e}steinn",
editor = "Charpentier, Lucas and
Choshen, Leshem and
Cotterell, Ryan and
Gul, Mustafa Omer and
Hu, Michael Y. and
Liu, Jing and
Jumelet, Jaap and
Linzen, Tal and
Mueller, Aaron and
Ross, Candace and
Shah, Raj Sanjay and
Warstadt, Alex and
Wilcox, Ethan Gotlieb and
Williams, Adina",
booktitle = "Proceedings of the First BabyLM Workshop",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.babylm-main.2/",
pages = "29--41",
ISBN = "TODO",
abstract = "Large language models (LLMs) are trained on huge amounts of textual data, and concerns have been raised that the limits of such data may soon be reached. A potential solution is to train on synthetic data sampled from LLMs. In this work, we build on this idea and investigate the benefits of *contrastive decoding* for generating synthetic data. In a controlled setting, we experiment with sampling corpora using the relative difference between a GOOD and BAD model trained on the same original corpus of 100 million words. By amplifying the signal from a model that has better performance, we create a synthetic corpus and mix it with the original training data. Our findings show that training on a mixture of synthesized and real data improves performance on the language modeling objective and a range of downstream tasks.In particular, we see that training with a mix of synthetic data from contrastive decoding benefits tasks that require more *reasoning skills*, while synthetic data from traditional sampling helps more on tasks requiring surface-level *linguistic* capabilities."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ulm-etal-2025-contrastive">
<titleInfo>
<title>Contrastive Decoding for Synthetic Data Generation in Low-Resource Language Modeling</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jannek</namePart>
<namePart type="family">Ulm</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kevin</namePart>
<namePart type="family">Du</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vésteinn</namePart>
<namePart type="family">Snæbjarnarson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the First BabyLM Workshop</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lucas</namePart>
<namePart type="family">Charpentier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Leshem</namePart>
<namePart type="family">Choshen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ryan</namePart>
<namePart type="family">Cotterell</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mustafa</namePart>
<namePart type="given">Omer</namePart>
<namePart type="family">Gul</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Michael</namePart>
<namePart type="given">Y</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaap</namePart>
<namePart type="family">Jumelet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tal</namePart>
<namePart type="family">Linzen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aaron</namePart>
<namePart type="family">Mueller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Candace</namePart>
<namePart type="family">Ross</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Raj</namePart>
<namePart type="given">Sanjay</namePart>
<namePart type="family">Shah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alex</namePart>
<namePart type="family">Warstadt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ethan</namePart>
<namePart type="given">Gotlieb</namePart>
<namePart type="family">Wilcox</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Adina</namePart>
<namePart type="family">Williams</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">TODO</identifier>
</relatedItem>
<abstract>Large language models (LLMs) are trained on huge amounts of textual data, and concerns have been raised that the limits of such data may soon be reached. A potential solution is to train on synthetic data sampled from LLMs. In this work, we build on this idea and investigate the benefits of *contrastive decoding* for generating synthetic data. In a controlled setting, we experiment with sampling corpora using the relative difference between a GOOD and BAD model trained on the same original corpus of 100 million words. By amplifying the signal from a model that has better performance, we create a synthetic corpus and mix it with the original training data. Our findings show that training on a mixture of synthesized and real data improves performance on the language modeling objective and a range of downstream tasks.In particular, we see that training with a mix of synthetic data from contrastive decoding benefits tasks that require more *reasoning skills*, while synthetic data from traditional sampling helps more on tasks requiring surface-level *linguistic* capabilities.</abstract>
<identifier type="citekey">ulm-etal-2025-contrastive</identifier>
<location>
<url>https://aclanthology.org/2025.babylm-main.2/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>29</start>
<end>41</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Contrastive Decoding for Synthetic Data Generation in Low-Resource Language Modeling
%A Ulm, Jannek
%A Du, Kevin
%A Snæbjarnarson, Vésteinn
%Y Charpentier, Lucas
%Y Choshen, Leshem
%Y Cotterell, Ryan
%Y Gul, Mustafa Omer
%Y Hu, Michael Y.
%Y Liu, Jing
%Y Jumelet, Jaap
%Y Linzen, Tal
%Y Mueller, Aaron
%Y Ross, Candace
%Y Shah, Raj Sanjay
%Y Warstadt, Alex
%Y Wilcox, Ethan Gotlieb
%Y Williams, Adina
%S Proceedings of the First BabyLM Workshop
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ TODO
%F ulm-etal-2025-contrastive
%X Large language models (LLMs) are trained on huge amounts of textual data, and concerns have been raised that the limits of such data may soon be reached. A potential solution is to train on synthetic data sampled from LLMs. In this work, we build on this idea and investigate the benefits of *contrastive decoding* for generating synthetic data. In a controlled setting, we experiment with sampling corpora using the relative difference between a GOOD and BAD model trained on the same original corpus of 100 million words. By amplifying the signal from a model that has better performance, we create a synthetic corpus and mix it with the original training data. Our findings show that training on a mixture of synthesized and real data improves performance on the language modeling objective and a range of downstream tasks.In particular, we see that training with a mix of synthetic data from contrastive decoding benefits tasks that require more *reasoning skills*, while synthetic data from traditional sampling helps more on tasks requiring surface-level *linguistic* capabilities.
%U https://aclanthology.org/2025.babylm-main.2/
%P 29-41
Markdown (Informal)
[Contrastive Decoding for Synthetic Data Generation in Low-Resource Language Modeling](https://aclanthology.org/2025.babylm-main.2/) (Ulm et al., BabyLM 2025)
ACL