Data Whisperer: Efficient Data Selection for Task-Specific LLM Fine-Tuning via Few-Shot In-Context Learning Article Swipe

PDF

Shaobo Wang , Xiaopeng Jin , Ziming Wang , Wang Jie , Jiajun Zhang , Kaixin Li , Zichen Wen , Zhong Li , Conghui He , Xuming Hu , Linfeng Zhang ·

YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2505.12212

Fine-tuning large language models (LLMs) on task-specific data is essential for their effective deployment. As dataset sizes grow, efficiently selecting optimal subsets for training becomes crucial to balancing performance and computational costs. Traditional data selection methods often require fine-tuning a scoring model on the target dataset, which is time-consuming and resource-intensive, or rely on heuristics that fail to fully leverage the model's predictive capabilities. To address these challenges, we propose Data Whisperer, an efficient, training-free, attention-based method that leverages few-shot in-context learning with the model to be fine-tuned. Comprehensive evaluations were conducted on both raw and synthetic datasets across diverse tasks and models. Notably, Data Whisperer achieves superior performance compared to the full GSM8K dataset on the Llama-3-8B-Instruct model, using just 10% of the data, and outperforms existing methods with a 3.1-point improvement and a 7.4$\times$ speedup. The code is available at https://github.com/gszfwsb/Data-Whisperer.

Related Topics

Computer Data Storage

Tokenization (Data Security)

Data Mart

International Data Group

Unstructured Supplementary Service Data

Data Storage

Data Fusion

Erwin Data Modeler

Concepts

No concepts available.

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2505.12212
PDF: https://arxiv.org/pdf/2505.12212
OA Status: green
OpenAlex ID: https://openalex.org/W4416445292

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4416445292

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2505.12212

Digital Object Identifier
Title: Data Whisperer: Efficient Data Selection for Task-Specific LLM Fine-Tuning via Few-Shot In-Context Learning

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2025

Year of publication
Publication date: 2025-05-18

Full publication date if available
Authors: Shaobo Wang, Xiaopeng Jin, Ziming Wang, Wang Jie, Jiajun Zhang, Kaixin Li, Zichen Wen, Zhong Li, Conghui He, Xuming Hu, Linfeng Zhang

List of authors in order
Landing page: https://arxiv.org/abs/2505.12212

Publisher landing page
PDF URL: https://arxiv.org/pdf/2505.12212

Direct link to full text PDF
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://arxiv.org/pdf/2505.12212

Direct OA link when available
Cited by: 0

Total citation count in OpenAlex

Full payload

id	https://openalex.org/W4416445292
doi	https://doi.org/10.48550/arxiv.2505.12212
ids.doi	https://doi.org/10.48550/arxiv.2505.12212
ids.openalex	https://openalex.org/W4416445292
fwci
type	preprint
title	Data Whisperer: Efficient Data Selection for Task-Specific LLM Fine-Tuning via Few-Shot In-Context Learning
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
is_xpac	False
apc_list
apc_paid
language	en
locations[0].id	pmh:oai:arXiv.org:2505.12212
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url	https://arxiv.org/pdf/2505.12212
locations[0].version	submittedVersion
locations[0].raw_type	text
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published	False
locations[0].raw_source_name
locations[0].landing_page_url	http://arxiv.org/abs/2505.12212
locations[1].id	doi:10.48550/arxiv.2505.12212
locations[1].is_oa	True
locations[1].source.id	https://openalex.org/S4306400194
locations[1].source.issn
locations[1].source.type	repository
locations[1].source.is_oa	True
locations[1].source.issn_l
locations[1].source.is_core	False
locations[1].source.is_in_doaj	False
locations[1].source.display_name	arXiv (Cornell University)
locations[1].source.host_organization	https://openalex.org/I205783295
locations[1].source.host_organization_name	Cornell University
locations[1].source.host_organization_lineage	https://openalex.org/I205783295
locations[1].license	cc-by
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id	https://openalex.org/licenses/cc-by
locations[1].is_accepted	False
locations[1].is_published
locations[1].raw_source_name
locations[1].landing_page_url	https://doi.org/10.48550/arxiv.2505.12212
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5107930049
authorships[0].author.orcid	https://orcid.org/0000-0001-8156-7081
authorships[0].author.display_name	Shaobo Wang
authorships[0].author_position	first
authorships[0].raw_author_name	Wang, Shaobo
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5002557354
authorships[1].author.orcid	https://orcid.org/0000-0002-6143-8570
authorships[1].author.display_name	Xiaopeng Jin
authorships[1].author_position	middle
authorships[1].raw_author_name	Jin, Xiangqi
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5100451885
authorships[2].author.orcid	https://orcid.org/0000-0002-5457-8882
authorships[2].author.display_name	Ziming Wang
authorships[2].author_position	middle
authorships[2].raw_author_name	Wang, Ziming
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5101828800
authorships[3].author.orcid	https://orcid.org/0000-0001-9704-3725
authorships[3].author.display_name	Wang Jie
authorships[3].author_position	middle
authorships[3].raw_author_name	Wang, Jize
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A5100319572
authorships[4].author.orcid	https://orcid.org/0000-0001-5293-7434
authorships[4].author.display_name	Jiajun Zhang
authorships[4].author_position	middle
authorships[4].raw_author_name	Zhang, Jiajun
authorships[4].is_corresponding	False
authorships[5].author.id	https://openalex.org/A5101812386
authorships[5].author.orcid	https://orcid.org/0000-0002-8876-6371
authorships[5].author.display_name	Kaixin Li
authorships[5].author_position	middle
authorships[5].raw_author_name	Li, Kaixin
authorships[5].is_corresponding	False
authorships[6].author.id	https://openalex.org/A5102635197
authorships[6].author.orcid	https://orcid.org/0009-0002-6157-5898
authorships[6].author.display_name	Zichen Wen
authorships[6].author_position	middle
authorships[6].raw_author_name	Wen, Zichen
authorships[6].is_corresponding	False
authorships[7].author.id	https://openalex.org/A5100664779
authorships[7].author.orcid	https://orcid.org/0000-0002-3849-3416
authorships[7].author.display_name	Zhong Li
authorships[7].author_position	middle
authorships[7].raw_author_name	Li, Zhong
authorships[7].is_corresponding	False
authorships[8].author.id	https://openalex.org/A5072088593
authorships[8].author.orcid	https://orcid.org/0009-0009-3186-6847
authorships[8].author.display_name	Conghui He
authorships[8].author_position	middle
authorships[8].raw_author_name	He, Conghui
authorships[8].is_corresponding	False
authorships[9].author.id	https://openalex.org/A5057914558
authorships[9].author.orcid	https://orcid.org/0000-0002-6973-8252
authorships[9].author.display_name	Xuming Hu
authorships[9].author_position	middle
authorships[9].raw_author_name	Hu, Xuming
authorships[9].is_corresponding	False
authorships[10].author.id	https://openalex.org/A5100689117
authorships[10].author.orcid	https://orcid.org/0000-0002-8470-5846
authorships[10].author.display_name	Linfeng Zhang
authorships[10].author_position	last
authorships[10].raw_author_name	Zhang, Linfeng
authorships[10].is_corresponding	False
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://arxiv.org/pdf/2505.12212
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-10-10T00:00:00
display_name	Data Whisperer: Efficient Data Selection for Task-Specific LLM Fine-Tuning via Few-Shot In-Context Learning
has_fulltext	False
is_retracted	False
updated_date	2025-12-01T00:03:43.161839
primary_topic
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2505.12212
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url	https://arxiv.org/pdf/2505.12212
best_oa_location.version	submittedVersion
best_oa_location.raw_type	text
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	http://arxiv.org/abs/2505.12212
primary_location.id	pmh:oai:arXiv.org:2505.12212
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url	https://arxiv.org/pdf/2505.12212
primary_location.version	submittedVersion
primary_location.raw_type	text
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	http://arxiv.org/abs/2505.12212
publication_date	2025-05-18
publication_year	2025
referenced_works_count	0
abstract_inverted_index.a	39, 130, 134
abstract_inverted_index.As	14
abstract_inverted_index.To	64
abstract_inverted_index.an	72
abstract_inverted_index.at	141
abstract_inverted_index.be	86
abstract_inverted_index.is	8, 47, 139
abstract_inverted_index.of	122
abstract_inverted_index.on	5, 42, 53, 92, 115
abstract_inverted_index.or	51
abstract_inverted_index.to	26, 57, 85, 110
abstract_inverted_index.we	68
abstract_inverted_index.10%	121
abstract_inverted_index.The	137
abstract_inverted_index.and	29, 49, 95, 101, 125, 133
abstract_inverted_index.for	10, 22
abstract_inverted_index.raw	94
abstract_inverted_index.the	43, 60, 83, 111, 116, 123
abstract_inverted_index.Data	70, 104
abstract_inverted_index.both	93
abstract_inverted_index.code	138
abstract_inverted_index.data	7, 33
abstract_inverted_index.fail	56
abstract_inverted_index.full	112
abstract_inverted_index.just	120
abstract_inverted_index.rely	52
abstract_inverted_index.that	55, 77
abstract_inverted_index.were	90
abstract_inverted_index.with	82, 129
abstract_inverted_index.GSM8K	113
abstract_inverted_index.data,	124
abstract_inverted_index.fully	58
abstract_inverted_index.grow,	17
abstract_inverted_index.large	1
abstract_inverted_index.model	41, 84
abstract_inverted_index.often	36
abstract_inverted_index.sizes	16
abstract_inverted_index.tasks	100
abstract_inverted_index.their	11
abstract_inverted_index.these	66
abstract_inverted_index.using	119
abstract_inverted_index.which	46
abstract_inverted_index.(LLMs)	4
abstract_inverted_index.across	98
abstract_inverted_index.costs.	31
abstract_inverted_index.method	76
abstract_inverted_index.model,	118
abstract_inverted_index.models	3
abstract_inverted_index.target	44
abstract_inverted_index.address	65
abstract_inverted_index.becomes	24
abstract_inverted_index.crucial	25
abstract_inverted_index.dataset	15, 114
abstract_inverted_index.diverse	99
abstract_inverted_index.methods	35, 128
abstract_inverted_index.model's	61
abstract_inverted_index.models.	102
abstract_inverted_index.optimal	20
abstract_inverted_index.propose	69
abstract_inverted_index.require	37
abstract_inverted_index.scoring	40
abstract_inverted_index.subsets	21
abstract_inverted_index.Notably,	103
abstract_inverted_index.achieves	106
abstract_inverted_index.compared	109
abstract_inverted_index.dataset,	45
abstract_inverted_index.datasets	97
abstract_inverted_index.existing	127
abstract_inverted_index.few-shot	79
abstract_inverted_index.language	2
abstract_inverted_index.learning	81
abstract_inverted_index.leverage	59
abstract_inverted_index.speedup.	136
abstract_inverted_index.superior	107
abstract_inverted_index.training	23
abstract_inverted_index.3.1-point	131
abstract_inverted_index.Whisperer	105
abstract_inverted_index.available	140
abstract_inverted_index.balancing	27
abstract_inverted_index.conducted	91
abstract_inverted_index.effective	12
abstract_inverted_index.essential	9
abstract_inverted_index.leverages	78
abstract_inverted_index.selecting	19
abstract_inverted_index.selection	34
abstract_inverted_index.synthetic	96
abstract_inverted_index.Whisperer,	71
abstract_inverted_index.efficient,	73
abstract_inverted_index.heuristics	54
abstract_inverted_index.in-context	80
abstract_inverted_index.predictive	62
abstract_inverted_index.7.4$\times$	135
abstract_inverted_index.Fine-tuning	0
abstract_inverted_index.Traditional	32
abstract_inverted_index.challenges,	67
abstract_inverted_index.deployment.	13
abstract_inverted_index.efficiently	18
abstract_inverted_index.evaluations	89
abstract_inverted_index.fine-tuned.	87
abstract_inverted_index.fine-tuning	38
abstract_inverted_index.improvement	132
abstract_inverted_index.outperforms	126
abstract_inverted_index.performance	28, 108
abstract_inverted_index.Comprehensive	88
abstract_inverted_index.capabilities.	63
abstract_inverted_index.computational	30
abstract_inverted_index.task-specific	6
abstract_inverted_index.time-consuming	48
abstract_inverted_index.training-free,	74
abstract_inverted_index.attention-based	75
abstract_inverted_index.Llama-3-8B-Instruct	117
abstract_inverted_index.resource-intensive,	50
abstract_inverted_index.https://github.com/gszfwsb/Data-Whisperer.	142
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	11
citation_normalized_percentile