View in EDS

Democratizing Advanced Attribution Analyses of Generative Language Models with the Inseq Toolkit

Saved in:
Bibliographic Details
Title: Democratizing Advanced Attribution Analyses of Generative Language Models with the Inseq Toolkit
Authors: Sarti, Gabriele, Feldhus, Nils, Qi, Jirui, Nissim, Malvina, Bisazza, Arianna
Publisher Information: CEUR Workshop Proceedings (CEUR-WS.org), 2024.
Publication Year: 2024
Subject Terms: Feature Attribution, Generative Language Models, Python Toolkit, Natural Language Processing
Description: Inseq1 is a recent toolkit providing an intuitive and optimized interface to conduct feature attribution analyses of generative language models. In this work, we present the latest improvements to the library, including efforts to simplify the attribution of large language models on consumer hardware, additional attribution approaches, and a new client command to detect and attribute context usage in language model generations. We showcase an online demo using Inseq as an attribution backbone for context reliance analysis, and we highlight interesting contextual patterns in language model generations. Ultimately, this release furthers Inseq’s mission of centralizing good interpretability practices and enabling fair and reproducible model evaluations.
Document Type: Conference object
Language: English
Access URL: https://research.rug.nl/en/publications/f719d93e-ca37-4965-b935-69bc53a48a4f
https://hdl.handle.net/11370/f719d93e-ca37-4965-b935-69bc53a48a4f
Rights: CC BY
Accession Number: edsair.dris...01423..3f1b08ba700891a86a4a076231d8432f
Database: OpenAIRE
FullText Text:
  Availability: 0
CustomLinks:
  – Url: https://explore.openaire.eu/search/publication?articleId=dris___01423%3A%3A3f1b08ba700891a86a4a076231d8432f
    Name: EDS - OpenAIRE (s4221598)
    Category: fullText
    Text: View record at OpenAIRE
  – Url: https://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=EBSCO&SrcAuth=EBSCO&DestApp=WOS&ServiceName=TransferToWoS&DestLinkType=GeneralSearchSummary&Func=Links&author=Sarti%20G
    Name: ISI
    Category: fullText
    Text: Nájsť tento článok vo Web of Science
    Icon: https://imagesrvr.epnet.com/ls/20docs.gif
    MouseOverText: Nájsť tento článok vo Web of Science
Header DbId: edsair
DbLabel: OpenAIRE
An: edsair.dris...01423..3f1b08ba700891a86a4a076231d8432f
RelevancyScore: 929
AccessLevel: 3
PubType: Conference
PubTypeId: conference
PreciseRelevancyScore: 929.415405273438
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Democratizing Advanced Attribution Analyses of Generative Language Models with the Inseq Toolkit
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Sarti%2C+Gabriele%22">Sarti, Gabriele</searchLink><br /><searchLink fieldCode="AR" term="%22Feldhus%2C+Nils%22">Feldhus, Nils</searchLink><br /><searchLink fieldCode="AR" term="%22Qi%2C+Jirui%22">Qi, Jirui</searchLink><br /><searchLink fieldCode="AR" term="%22Nissim%2C+Malvina%22">Nissim, Malvina</searchLink><br /><searchLink fieldCode="AR" term="%22Bisazza%2C+Arianna%22">Bisazza, Arianna</searchLink>
– Name: Publisher
  Label: Publisher Information
  Group: PubInfo
  Data: CEUR Workshop Proceedings (CEUR-WS.org), 2024.
– Name: DatePubCY
  Label: Publication Year
  Group: Date
  Data: 2024
– Name: Subject
  Label: Subject Terms
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Feature+Attribution%22">Feature Attribution</searchLink><br /><searchLink fieldCode="DE" term="%22Generative+Language+Models%22">Generative Language Models</searchLink><br /><searchLink fieldCode="DE" term="%22Python+Toolkit%22">Python Toolkit</searchLink><br /><searchLink fieldCode="DE" term="%22Natural+Language+Processing%22">Natural Language Processing</searchLink>
– Name: Abstract
  Label: Description
  Group: Ab
  Data: Inseq1 is a recent toolkit providing an intuitive and optimized interface to conduct feature attribution analyses of generative language models. In this work, we present the latest improvements to the library, including efforts to simplify the attribution of large language models on consumer hardware, additional attribution approaches, and a new client command to detect and attribute context usage in language model generations. We showcase an online demo using Inseq as an attribution backbone for context reliance analysis, and we highlight interesting contextual patterns in language model generations. Ultimately, this release furthers Inseq’s mission of centralizing good interpretability practices and enabling fair and reproducible model evaluations.
– Name: TypeDocument
  Label: Document Type
  Group: TypDoc
  Data: Conference object
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: URL
  Label: Access URL
  Group: URL
  Data: <link linkTarget="URL" linkTerm="https://research.rug.nl/en/publications/f719d93e-ca37-4965-b935-69bc53a48a4f" linkWindow="_blank">https://research.rug.nl/en/publications/f719d93e-ca37-4965-b935-69bc53a48a4f</link><br /><link linkTarget="URL" linkTerm="https://hdl.handle.net/11370/f719d93e-ca37-4965-b935-69bc53a48a4f" linkWindow="_blank">https://hdl.handle.net/11370/f719d93e-ca37-4965-b935-69bc53a48a4f</link>
– Name: Copyright
  Label: Rights
  Group: Cpyrght
  Data: CC BY
– Name: AN
  Label: Accession Number
  Group: ID
  Data: edsair.dris...01423..3f1b08ba700891a86a4a076231d8432f
PLink https://erproxy.cvtisr.sk/sfx/access?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsair&AN=edsair.dris...01423..3f1b08ba700891a86a4a076231d8432f
RecordInfo BibRecord:
  BibEntity:
    Languages:
      – Text: English
    Subjects:
      – SubjectFull: Feature Attribution
        Type: general
      – SubjectFull: Generative Language Models
        Type: general
      – SubjectFull: Python Toolkit
        Type: general
      – SubjectFull: Natural Language Processing
        Type: general
    Titles:
      – TitleFull: Democratizing Advanced Attribution Analyses of Generative Language Models with the Inseq Toolkit
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Sarti, Gabriele
      – PersonEntity:
          Name:
            NameFull: Feldhus, Nils
      – PersonEntity:
          Name:
            NameFull: Qi, Jirui
      – PersonEntity:
          Name:
            NameFull: Nissim, Malvina
      – PersonEntity:
          Name:
            NameFull: Bisazza, Arianna
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 01
              Type: published
              Y: 2024
          Identifiers:
            – Type: issn-locals
              Value: edsair
            – Type: issn-locals
              Value: edsairFT
ResultId 1