@inproceedings{kim-etal-2025-benchmark,
title = "Benchmark Profiling: Mechanistic Diagnosis of {LLM} Benchmarks",
author = "Kim, Dongjun and
Shim, Gyuho and
Chun, Yongchan and
Kim, Minhyuk and
Park, Chanjun and
Lim, Heuiseok",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.emnlp-main.789/",
doi = "10.18653/v1/2025.emnlp-main.789",
pages = "15635--15650",
ISBN = "979-8-89176-332-6",
abstract = "Large Language Models are commonly judged by their scores on standard benchmarks, yet such scores often overstate real capability since they mask the mix of skills a task actually demands. For example, ARC is assumed to test reasoning, while HellaSwag is designed to evaluate commonsense. However, we lack a systematic way to verify if these benchmarks actually measure these labels. We introduce **BENCHMARK PROFILING**, a diagnostic framework that decomposes benchmark performance into ten cognitively grounded abilities. The method combines gradient-based importance scoring with targeted parameter ablation to compute an Ability Impact Score (AIS) that quantifies how much each ability contributes to a model{'}s success on a given benchmark. Profiling three instruction-tuned models across ten widely used benchmarks yields four key findings: (i) most benchmarks draw on several abilities rather than one, (ii) datasets with similar labels rely on distinct ability mixtures, (iii) code-generation benchmarks reward broad, multi-skill improvement and thus show only modest gains from narrow domain-specific fine-tuning, and (iv) abilities irrelevant to the task could negatively affect performance. **BENCHMARK PROFILING** therefore explains why performance gains do not always translate into user-perceived competence and offer a transparent tool for benchmark audit and model interpretability."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kim-etal-2025-benchmark">
<titleInfo>
<title>Benchmark Profiling: Mechanistic Diagnosis of LLM Benchmarks</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dongjun</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gyuho</namePart>
<namePart type="family">Shim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yongchan</namePart>
<namePart type="family">Chun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minhyuk</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chanjun</namePart>
<namePart type="family">Park</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Heuiseok</namePart>
<namePart type="family">Lim</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 2025 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-332-6</identifier>
</relatedItem>
<abstract>Large Language Models are commonly judged by their scores on standard benchmarks, yet such scores often overstate real capability since they mask the mix of skills a task actually demands. For example, ARC is assumed to test reasoning, while HellaSwag is designed to evaluate commonsense. However, we lack a systematic way to verify if these benchmarks actually measure these labels. We introduce **BENCHMARK PROFILING**, a diagnostic framework that decomposes benchmark performance into ten cognitively grounded abilities. The method combines gradient-based importance scoring with targeted parameter ablation to compute an Ability Impact Score (AIS) that quantifies how much each ability contributes to a model’s success on a given benchmark. Profiling three instruction-tuned models across ten widely used benchmarks yields four key findings: (i) most benchmarks draw on several abilities rather than one, (ii) datasets with similar labels rely on distinct ability mixtures, (iii) code-generation benchmarks reward broad, multi-skill improvement and thus show only modest gains from narrow domain-specific fine-tuning, and (iv) abilities irrelevant to the task could negatively affect performance. **BENCHMARK PROFILING** therefore explains why performance gains do not always translate into user-perceived competence and offer a transparent tool for benchmark audit and model interpretability.</abstract>
<identifier type="citekey">kim-etal-2025-benchmark</identifier>
<identifier type="doi">10.18653/v1/2025.emnlp-main.789</identifier>
<location>
<url>https://aclanthology.org/2025.emnlp-main.789/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>15635</start>
<end>15650</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Benchmark Profiling: Mechanistic Diagnosis of LLM Benchmarks
%A Kim, Dongjun
%A Shim, Gyuho
%A Chun, Yongchan
%A Kim, Minhyuk
%A Park, Chanjun
%A Lim, Heuiseok
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-332-6
%F kim-etal-2025-benchmark
%X Large Language Models are commonly judged by their scores on standard benchmarks, yet such scores often overstate real capability since they mask the mix of skills a task actually demands. For example, ARC is assumed to test reasoning, while HellaSwag is designed to evaluate commonsense. However, we lack a systematic way to verify if these benchmarks actually measure these labels. We introduce **BENCHMARK PROFILING**, a diagnostic framework that decomposes benchmark performance into ten cognitively grounded abilities. The method combines gradient-based importance scoring with targeted parameter ablation to compute an Ability Impact Score (AIS) that quantifies how much each ability contributes to a model’s success on a given benchmark. Profiling three instruction-tuned models across ten widely used benchmarks yields four key findings: (i) most benchmarks draw on several abilities rather than one, (ii) datasets with similar labels rely on distinct ability mixtures, (iii) code-generation benchmarks reward broad, multi-skill improvement and thus show only modest gains from narrow domain-specific fine-tuning, and (iv) abilities irrelevant to the task could negatively affect performance. **BENCHMARK PROFILING** therefore explains why performance gains do not always translate into user-perceived competence and offer a transparent tool for benchmark audit and model interpretability.
%R 10.18653/v1/2025.emnlp-main.789
%U https://aclanthology.org/2025.emnlp-main.789/
%U https://doi.org/10.18653/v1/2025.emnlp-main.789
%P 15635-15650
Markdown (Informal)
[Benchmark Profiling: Mechanistic Diagnosis of LLM Benchmarks](https://aclanthology.org/2025.emnlp-main.789/) (Kim et al., EMNLP 2025)
ACL
- Dongjun Kim, Gyuho Shim, Yongchan Chun, Minhyuk Kim, Chanjun Park, and Heuiseok Lim. 2025. Benchmark Profiling: Mechanistic Diagnosis of LLM Benchmarks. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 15635–15650, Suzhou, China. Association for Computational Linguistics.