prikaz prve stranice dokumenta Kreiranje fiktivnih priča koristeći GPT-Neo
  Download
PDF 742.98 KB
undergraduate thesis
Kreiranje fiktivnih priča koristeći GPT-Neo
Pula: University of Pula, 2023. urn:nbn:hr:137:110695
Kanjovsky, Tin
University of Pula
Faculty of Informatics in Pula

Institutional repository: Digital Repository Juraj Dobrila University of Pula

Cite this document

Kanjovsky, T. (2023). Kreiranje fiktivnih priča koristeći GPT-Neo (Undergraduate thesis). Pula: University of Pula. Retrieved from https://urn.nsk.hr/urn:nbn:hr:137:110695

Kanjovsky, Tin. "Kreiranje fiktivnih priča koristeći GPT-Neo." Undergraduate thesis, University of Pula, 2023. https://urn.nsk.hr/urn:nbn:hr:137:110695

Kanjovsky, Tin. "Kreiranje fiktivnih priča koristeći GPT-Neo." Undergraduate thesis, University of Pula, 2023. https://urn.nsk.hr/urn:nbn:hr:137:110695

Kanjovsky, T. (2023). 'Kreiranje fiktivnih priča koristeći GPT-Neo', Undergraduate thesis, University of Pula, accessed 06 May 2024, https://urn.nsk.hr/urn:nbn:hr:137:110695

Kanjovsky T. Kreiranje fiktivnih priča koristeći GPT-Neo [Undergraduate thesis]. Pula: University of Pula; 2023 [cited 2024 May 06] Available at: https://urn.nsk.hr/urn:nbn:hr:137:110695

T. Kanjovsky, "Kreiranje fiktivnih priča koristeći GPT-Neo", Undergraduate thesis, University of Pula, Pula, 2023. Available at: https://urn.nsk.hr/urn:nbn:hr:137:110695

Metadata
TitleKreiranje fiktivnih priča koristeći GPT-Neo
Title (english)Making a fiction story generator using a GPT-Neo
AuthorTin Kanjovsky
MentorDarko Etinger (mentor)
Committee memberSnježana Babić (predsjednik povjerenstva)
Committee memberRomeo Šajina (član povjerenstva)
Committee memberDarko Etinger (član povjerenstva)
GranterUniversity of Pula
(Faculty of Informatics in Pula)
Pula
Defense date and country2023-09-29, Croatia
Scientific / art field,
discipline and subdiscipline		SOCIAL SCIENCES
Information and Communication Sciences
Information Systems and Information Science
Abstract
Rad istražuje koncept generiranja izmišljenih priča koristeći GPT-Neo, besplatan i open-source jezični model. Generator bi imao za cilj olakšati stvaranje zanimljivih i originalnih izmišljenih priča. GPT-Neo-ova sposobnost učenja iz velikih skupova podataka i generiranja koherentnog i prirodnog jezika čini ga alatom koji bi mogao uspješno obaviti ovaj zadatak. Model je fino podešen koristeći veliki skupa podataka od 300 tisuća napisanih priča uparenih s upitima s online foruma. Suočavajući se s izazovom fino podešavanja GPT-Neo modela na posebno odabranom skupu podataka bogatom fikcijom, ostvareno je značajno poboljšanje u razumijevanju i kvaliteti generiranih tekstova. Fino podešeni model pokazuje značajna poboljšanja, kako i u metrikama poput perpleksnost, tako i u generiranim pričama u odnosu na osnovni model.
Abstract (english)
This paper explores the concept of generating fictional stories using GPT-Neo, a free and open-source language model. The fiction story generator aims to facilitate the creation of interesting and original fictional stories. GPT-Neo's ability to learn from large datasets and generate coherent and natural language makes it a tool that could successfully perform this task. The model is fine-tuned using a large dataset of 300,000 written stories paired with prompts from online forums. Faced with the challenge of fine-tuning the GPT-Neo model on a specially selected dataset rich in fiction, significant improvements in the understanding and quality of generated texts have been achieved. The finely tuned model shows significant improvements both in metrics like perplexity and in the generated stories compared to the baseline model.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:137:110695
Study programmeTitle: Informatics
Study programme type: university
Study level: undergraduate
Academic / professional title: sveučilišni/a prvostupnik/prvostupnica (baccalaureus/baccalaurea) informatike (sveučilišni/a prvostupnik/prvostupnica (baccalaureus/baccalaurea) informatike)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2023-09-26 11:09:10