@inproceedings{ren-etal-2025-efficiently,
title = "Efficiently Selecting Response Generation Strategies for Synthetic Data Construction by Self-Aligned Perplexity",
author = "Ren, Xuan and
Chen, Qi and
Liu, Lingqiao",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-emnlp.621/",
pages = "11584--11605",
ISBN = "979-8-89176-335-7",
abstract = "Fine-tuning large language models (LLMs) typically relies on producing large sets of input-output pairs. Yet for a given question, there can be many valid outputs. In practice, these outputs are often derived by distilling knowledge from teacher models, and they can vary depending on the specific teacher model or prompting strategy employed.Recent findings show that how these training outputs are generated can significantly affect the performance of the fine-tuned model, raising an important question: how do we pick the best data generation method from among numerous possibilities? Rather than exhaustively training and evaluating on each candidate, this paper proposes a scalable approximate method that assesses a small subset of generated data to estimate its suitability for a specific target LLM. Our central idea is that effective outputs should be familiar to the target LLM. While previous work measures familiarity with perplexity, we find that perplexity might be suboptimal in characterizing ``familiarity'' through empirical analyses and practical observations. To address this, we introduce self-aligned perplexity, a novel metric capturing how closely candidate outputs adhere to the target LLM{'}s own style and reasoning patterns. In this way, we can identify the most effective generation strategy on a small sample, then apply it to produce the complete training set. We demonstrate that training on data generated by the chosen method yields significant improvements across diverse reasoning-focused benchmarks, particularly in cases where different candidate methods lead to highly divergent training outcomes."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ren-etal-2025-efficiently">
<titleInfo>
<title>Efficiently Selecting Response Generation Strategies for Synthetic Data Construction by Self-Aligned Perplexity</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xuan</namePart>
<namePart type="family">Ren</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lingqiao</namePart>
<namePart type="family">Liu</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>Findings of the Association for Computational Linguistics: EMNLP 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolyn</namePart>
<namePart type="family">Rose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Violet</namePart>
<namePart type="family">Peng</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">979-8-89176-335-7</identifier>
</relatedItem>
<abstract>Fine-tuning large language models (LLMs) typically relies on producing large sets of input-output pairs. Yet for a given question, there can be many valid outputs. In practice, these outputs are often derived by distilling knowledge from teacher models, and they can vary depending on the specific teacher model or prompting strategy employed.Recent findings show that how these training outputs are generated can significantly affect the performance of the fine-tuned model, raising an important question: how do we pick the best data generation method from among numerous possibilities? Rather than exhaustively training and evaluating on each candidate, this paper proposes a scalable approximate method that assesses a small subset of generated data to estimate its suitability for a specific target LLM. Our central idea is that effective outputs should be familiar to the target LLM. While previous work measures familiarity with perplexity, we find that perplexity might be suboptimal in characterizing “familiarity” through empirical analyses and practical observations. To address this, we introduce self-aligned perplexity, a novel metric capturing how closely candidate outputs adhere to the target LLM’s own style and reasoning patterns. In this way, we can identify the most effective generation strategy on a small sample, then apply it to produce the complete training set. We demonstrate that training on data generated by the chosen method yields significant improvements across diverse reasoning-focused benchmarks, particularly in cases where different candidate methods lead to highly divergent training outcomes.</abstract>
<identifier type="citekey">ren-etal-2025-efficiently</identifier>
<location>
<url>https://aclanthology.org/2025.findings-emnlp.621/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>11584</start>
<end>11605</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Efficiently Selecting Response Generation Strategies for Synthetic Data Construction by Self-Aligned Perplexity
%A Ren, Xuan
%A Chen, Qi
%A Liu, Lingqiao
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Findings of the Association for Computational Linguistics: EMNLP 2025
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-335-7
%F ren-etal-2025-efficiently
%X Fine-tuning large language models (LLMs) typically relies on producing large sets of input-output pairs. Yet for a given question, there can be many valid outputs. In practice, these outputs are often derived by distilling knowledge from teacher models, and they can vary depending on the specific teacher model or prompting strategy employed.Recent findings show that how these training outputs are generated can significantly affect the performance of the fine-tuned model, raising an important question: how do we pick the best data generation method from among numerous possibilities? Rather than exhaustively training and evaluating on each candidate, this paper proposes a scalable approximate method that assesses a small subset of generated data to estimate its suitability for a specific target LLM. Our central idea is that effective outputs should be familiar to the target LLM. While previous work measures familiarity with perplexity, we find that perplexity might be suboptimal in characterizing “familiarity” through empirical analyses and practical observations. To address this, we introduce self-aligned perplexity, a novel metric capturing how closely candidate outputs adhere to the target LLM’s own style and reasoning patterns. In this way, we can identify the most effective generation strategy on a small sample, then apply it to produce the complete training set. We demonstrate that training on data generated by the chosen method yields significant improvements across diverse reasoning-focused benchmarks, particularly in cases where different candidate methods lead to highly divergent training outcomes.
%U https://aclanthology.org/2025.findings-emnlp.621/
%P 11584-11605
Markdown (Informal)
[Efficiently Selecting Response Generation Strategies for Synthetic Data Construction by Self-Aligned Perplexity](https://aclanthology.org/2025.findings-emnlp.621/) (Ren et al., Findings 2025)
ACL