diplomsko delo
Robert Samardžija (Author), Zoran Bosnić (Mentor), Dragoslav Radin (Co-mentor)

Abstract

Področje generativne umetne inteligence je v letu 2022 v tehnološko stroko in tudi v ostale stroke prineslo revolucijo. Razcvet na področju osnovnih modelov je omogočil ustvarjanje realističnih in kompleksnih vsebin različnih vrst ter odprl vrata novim pristopom na področjih ustvarjalnosti, strojnega prevajanja in odločanja. V diplomski nalogi raziščemo uporabo velikih jezikovnih modelov za generiranje dokumentacije iz izvorne kode. Ogledamo si pristope inženiringa poizvedb, zasnujemo in razvijemo prototip generatorja ter ocenimo zmogljivost velikih jezikovnih modelov na zastavljeni nalogi. Izpostavimo težavo narave delovanja jezikovnih modelov, ki lahko pri različnih izvajanjih ustvarijo nezaželene razlike v rezultatih, in problem prilagajanja naše metode na delovanje specifičnega jezikovnega modela. Delo zaključimo z ugotovitvijo, da implementacija naše metode zadovoljuje potrebe podjetja DevRev in predstavlja alternativo obstoječim generatorjem dokumentacije, ki ne uporabljajo jezikovnih modelov. Predstavimo možne izboljšave, ki vključujejo uporabo jezikovnih modelov iz različnih družin in integracijo prototipa v storitev Airdrop platforme DevRev.

Keywords

veliki jezikovni modeli;generatorji dokumentacije; inženiring poizvedb;GPT;univerzitetni študij;diplomske naloge;

Data

Language: Slovenian
Year of publishing:
Typology: 2.11 - Undergraduate Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [R. Samardžija]
UDC: 004.85:81'322(043.2)
COBISS: 207569923 Link will open in a new window
Views: 239
Downloads: 66
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Use of large language models for generating source code documentation
Secondary abstract: The field of generative artificial intelligence brought about a revolution in technology and other disciplines in the year 2022. The development and incredible success of foundation models enabled the creation of realistic and complex content of various kinds and introduced new approaches in creativity, machine translation and decision-making. In our work, we explore the use of large language models for generating source code documentation. We examine prompt engineering approaches, design and develop a prototype of the generator and evaluate the performance of large language models on the set task. We highlight the challenging nature of language models, whose output can undesirably differ between runs, and the problem of tuning our method to one specific language model. The work concludes with the finding that the implementation of our method satisfies the needs of DevRev and represents an alternative to existing documentation generators that do not use language models. We also present possible improvements that include the use of language models from different families and the integration of our prototype into DevRev's Airdrop service.
Secondary keywords: large language models;documentation generators;prompt engineering;GPT;computer and information science;diploma;Računalniško jezikoslovje;Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS): Bachelor thesis/paper
Study programme: 1000468
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 49 str.
ID: 24676131