An Experimental Design Framework for Label-Efficient Supervised Finetuning of Large Language Models

Supervised finetuning (SFT) on instruction datasets has played a crucial role in achieving the remarkable zero-shot generalization capabilities observed in modern large language models (LLMs). However, the annotation efforts required to produce high quality responses for instructions are becoming pr...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Bhatt, Gantavya, Chen, Yifang, Das, Arnav M, Zhang, Jifan, Truong, Sang T, Mussmann, Stephen, Zhu, Yinglun, Bilmes, Jeffrey, Du, Simon S, Jamieson, Kevin, Ash, Jordan T, Nowak, Robert D
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Bhatt, Gantavya
Chen, Yifang
Das, Arnav M
Zhang, Jifan
Truong, Sang T
Mussmann, Stephen
Zhu, Yinglun
Bilmes, Jeffrey
Du, Simon S
Jamieson, Kevin
Ash, Jordan T
Nowak, Robert D
description Supervised finetuning (SFT) on instruction datasets has played a crucial role in achieving the remarkable zero-shot generalization capabilities observed in modern large language models (LLMs). However, the annotation efforts required to produce high quality responses for instructions are becoming prohibitively expensive, especially as the number of tasks spanned by instruction datasets continues to increase. Active learning is effective in identifying useful subsets of samples to annotate from an unlabeled pool, but its high computational cost remains a barrier to its widespread applicability in the context of LLMs. To mitigate the annotation cost of SFT and circumvent the computational bottlenecks of active learning, we propose using experimental design. Experimental design techniques select the most informative samples to label, and typically maximize some notion of uncertainty and/or diversity. In our work, we implement a framework that evaluates several existing and novel experimental design techniques and find that these methods consistently yield significant gains in label efficiency with little computational overhead. On generative tasks, our methods achieve the same generalization performance with only $50\%$ of annotation cost required by random sampling.
doi_str_mv 10.48550/arxiv.2401.06692
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2401_06692</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2401_06692</sourcerecordid><originalsourceid>FETCH-LOGICAL-a672-76f057061117ee41ce4e045f867fb859d47133a5f322202083e71b2417aba253</originalsourceid><addsrcrecordid>eNotj79OwzAYxL0woMIDMOEXSPB_p2NVEkBKxVD26EvyObJIncppSnl7TGG5u-HupB8hD5zlqtCaPUG8-HMuFOM5M2Ytbkm3CbS8HDH6A4YTjPQZZz8EWkU44NcUP6mbIq2hxTErnfOdTzW6X9Li7GfsaeUDnpbgw0Anl4pxwKRhWCCF3dTjON-RGwfjjPf_viL7qvzYvmb1-8vbdlNnYKzIrHFMW2Y45xZR8Q4VMqVdYaxrC73uleVSgnZSCMEEKyRa3grFLbQgtFyRx7_XK2RzTEQQv5tf2OYKK38AK3xOwQ</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>An Experimental Design Framework for Label-Efficient Supervised Finetuning of Large Language Models</title><source>arXiv.org</source><creator>Bhatt, Gantavya ; Chen, Yifang ; Das, Arnav M ; Zhang, Jifan ; Truong, Sang T ; Mussmann, Stephen ; Zhu, Yinglun ; Bilmes, Jeffrey ; Du, Simon S ; Jamieson, Kevin ; Ash, Jordan T ; Nowak, Robert D</creator><creatorcontrib>Bhatt, Gantavya ; Chen, Yifang ; Das, Arnav M ; Zhang, Jifan ; Truong, Sang T ; Mussmann, Stephen ; Zhu, Yinglun ; Bilmes, Jeffrey ; Du, Simon S ; Jamieson, Kevin ; Ash, Jordan T ; Nowak, Robert D</creatorcontrib><description>Supervised finetuning (SFT) on instruction datasets has played a crucial role in achieving the remarkable zero-shot generalization capabilities observed in modern large language models (LLMs). However, the annotation efforts required to produce high quality responses for instructions are becoming prohibitively expensive, especially as the number of tasks spanned by instruction datasets continues to increase. Active learning is effective in identifying useful subsets of samples to annotate from an unlabeled pool, but its high computational cost remains a barrier to its widespread applicability in the context of LLMs. To mitigate the annotation cost of SFT and circumvent the computational bottlenecks of active learning, we propose using experimental design. Experimental design techniques select the most informative samples to label, and typically maximize some notion of uncertainty and/or diversity. In our work, we implement a framework that evaluates several existing and novel experimental design techniques and find that these methods consistently yield significant gains in label efficiency with little computational overhead. On generative tasks, our methods achieve the same generalization performance with only $50\%$ of annotation cost required by random sampling.</description><identifier>DOI: 10.48550/arxiv.2401.06692</identifier><language>eng</language><subject>Computer Science - Artificial Intelligence ; Computer Science - Computation and Language ; Computer Science - Learning</subject><creationdate>2024-01</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2401.06692$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2401.06692$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Bhatt, Gantavya</creatorcontrib><creatorcontrib>Chen, Yifang</creatorcontrib><creatorcontrib>Das, Arnav M</creatorcontrib><creatorcontrib>Zhang, Jifan</creatorcontrib><creatorcontrib>Truong, Sang T</creatorcontrib><creatorcontrib>Mussmann, Stephen</creatorcontrib><creatorcontrib>Zhu, Yinglun</creatorcontrib><creatorcontrib>Bilmes, Jeffrey</creatorcontrib><creatorcontrib>Du, Simon S</creatorcontrib><creatorcontrib>Jamieson, Kevin</creatorcontrib><creatorcontrib>Ash, Jordan T</creatorcontrib><creatorcontrib>Nowak, Robert D</creatorcontrib><title>An Experimental Design Framework for Label-Efficient Supervised Finetuning of Large Language Models</title><description>Supervised finetuning (SFT) on instruction datasets has played a crucial role in achieving the remarkable zero-shot generalization capabilities observed in modern large language models (LLMs). However, the annotation efforts required to produce high quality responses for instructions are becoming prohibitively expensive, especially as the number of tasks spanned by instruction datasets continues to increase. Active learning is effective in identifying useful subsets of samples to annotate from an unlabeled pool, but its high computational cost remains a barrier to its widespread applicability in the context of LLMs. To mitigate the annotation cost of SFT and circumvent the computational bottlenecks of active learning, we propose using experimental design. Experimental design techniques select the most informative samples to label, and typically maximize some notion of uncertainty and/or diversity. In our work, we implement a framework that evaluates several existing and novel experimental design techniques and find that these methods consistently yield significant gains in label efficiency with little computational overhead. On generative tasks, our methods achieve the same generalization performance with only $50\%$ of annotation cost required by random sampling.</description><subject>Computer Science - Artificial Intelligence</subject><subject>Computer Science - Computation and Language</subject><subject>Computer Science - Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj79OwzAYxL0woMIDMOEXSPB_p2NVEkBKxVD26EvyObJIncppSnl7TGG5u-HupB8hD5zlqtCaPUG8-HMuFOM5M2Ytbkm3CbS8HDH6A4YTjPQZZz8EWkU44NcUP6mbIq2hxTErnfOdTzW6X9Li7GfsaeUDnpbgw0Anl4pxwKRhWCCF3dTjON-RGwfjjPf_viL7qvzYvmb1-8vbdlNnYKzIrHFMW2Y45xZR8Q4VMqVdYaxrC73uleVSgnZSCMEEKyRa3grFLbQgtFyRx7_XK2RzTEQQv5tf2OYKK38AK3xOwQ</recordid><startdate>20240112</startdate><enddate>20240112</enddate><creator>Bhatt, Gantavya</creator><creator>Chen, Yifang</creator><creator>Das, Arnav M</creator><creator>Zhang, Jifan</creator><creator>Truong, Sang T</creator><creator>Mussmann, Stephen</creator><creator>Zhu, Yinglun</creator><creator>Bilmes, Jeffrey</creator><creator>Du, Simon S</creator><creator>Jamieson, Kevin</creator><creator>Ash, Jordan T</creator><creator>Nowak, Robert D</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20240112</creationdate><title>An Experimental Design Framework for Label-Efficient Supervised Finetuning of Large Language Models</title><author>Bhatt, Gantavya ; Chen, Yifang ; Das, Arnav M ; Zhang, Jifan ; Truong, Sang T ; Mussmann, Stephen ; Zhu, Yinglun ; Bilmes, Jeffrey ; Du, Simon S ; Jamieson, Kevin ; Ash, Jordan T ; Nowak, Robert D</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a672-76f057061117ee41ce4e045f867fb859d47133a5f322202083e71b2417aba253</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Artificial Intelligence</topic><topic>Computer Science - Computation and Language</topic><topic>Computer Science - Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Bhatt, Gantavya</creatorcontrib><creatorcontrib>Chen, Yifang</creatorcontrib><creatorcontrib>Das, Arnav M</creatorcontrib><creatorcontrib>Zhang, Jifan</creatorcontrib><creatorcontrib>Truong, Sang T</creatorcontrib><creatorcontrib>Mussmann, Stephen</creatorcontrib><creatorcontrib>Zhu, Yinglun</creatorcontrib><creatorcontrib>Bilmes, Jeffrey</creatorcontrib><creatorcontrib>Du, Simon S</creatorcontrib><creatorcontrib>Jamieson, Kevin</creatorcontrib><creatorcontrib>Ash, Jordan T</creatorcontrib><creatorcontrib>Nowak, Robert D</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Bhatt, Gantavya</au><au>Chen, Yifang</au><au>Das, Arnav M</au><au>Zhang, Jifan</au><au>Truong, Sang T</au><au>Mussmann, Stephen</au><au>Zhu, Yinglun</au><au>Bilmes, Jeffrey</au><au>Du, Simon S</au><au>Jamieson, Kevin</au><au>Ash, Jordan T</au><au>Nowak, Robert D</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>An Experimental Design Framework for Label-Efficient Supervised Finetuning of Large Language Models</atitle><date>2024-01-12</date><risdate>2024</risdate><abstract>Supervised finetuning (SFT) on instruction datasets has played a crucial role in achieving the remarkable zero-shot generalization capabilities observed in modern large language models (LLMs). However, the annotation efforts required to produce high quality responses for instructions are becoming prohibitively expensive, especially as the number of tasks spanned by instruction datasets continues to increase. Active learning is effective in identifying useful subsets of samples to annotate from an unlabeled pool, but its high computational cost remains a barrier to its widespread applicability in the context of LLMs. To mitigate the annotation cost of SFT and circumvent the computational bottlenecks of active learning, we propose using experimental design. Experimental design techniques select the most informative samples to label, and typically maximize some notion of uncertainty and/or diversity. In our work, we implement a framework that evaluates several existing and novel experimental design techniques and find that these methods consistently yield significant gains in label efficiency with little computational overhead. On generative tasks, our methods achieve the same generalization performance with only $50\%$ of annotation cost required by random sampling.</abstract><doi>10.48550/arxiv.2401.06692</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2401.06692
ispartof
issn
language eng
recordid cdi_arxiv_primary_2401_06692
source arXiv.org
subjects Computer Science - Artificial Intelligence
Computer Science - Computation and Language
Computer Science - Learning
title An Experimental Design Framework for Label-Efficient Supervised Finetuning of Large Language Models
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-02T20%3A51%3A14IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=An%20Experimental%20Design%20Framework%20for%20Label-Efficient%20Supervised%20Finetuning%20of%20Large%20Language%20Models&rft.au=Bhatt,%20Gantavya&rft.date=2024-01-12&rft_id=info:doi/10.48550/arxiv.2401.06692&rft_dat=%3Carxiv_GOX%3E2401_06692%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true