Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/148726
Title: Entrenament mitjançant aprenentatge per reforç d’un model de llenguatge per a la generació automatitzada d’aplicacions
Other Titles: Entrenament d’un gran model de llenguatge per a la generació automatitzada d’aplicacions
Author: Masagué Deu, Quer
Tutor: Ferrer-Mestres, Jonathan  
Others: Baró, Xavier  
Abstract: This work aims to define a strategy to take advantage of the large language models for automated application generation. To this end, a generative language model is designed and trained using proprietary sources. While the use of these models in application development is currently limited to the context of assistance, their improving quality is making them increasingly suitable for automating tasks of this type. Although the privatization of access to pre-trained large models there is a large community working on open versions. It is proposed to use one of these open architectures, the NanoGPT, to train a model for this purpose. Due to the high computational cost and the large volumes of data required, the original datasets had to be multiplied using templates. In an iterative process, different configurations of the models have been trained and compared, seeking to improve the quality of their results. Through this approach and the application of prompt engineering techniques, the goal of generating small applications in an automated way with the required functionalities and parameters has been achieved. Applying these results, it becomes possible to train a model based on production code to provide support for an application that facilitates the automated creation of applications. Considering the cost of the necessary infrastructure, using a pre-trained model refined with custom code becomes an attractive option. This tool can be directly queried using natural language to obtain the required program without the need for an intermediary application.
Keywords: machine learning
natural language processing
large language model
Document type: info:eu-repo/semantics/bachelorThesis
Issue Date: Jun-2023
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
quermdTFG0623memoria.pdfMermòria del TFG3,12 MBAdobe PDFThumbnail
View/Open
quermdTFG0623videopresentacio.mkvVídeo de la presentació90,47 MBMKVView/Open
quermdTFG0623presentacio.odpPresentació16,53 MBOpenDocument PresentationView/Open
Share:
Export:
View statistics

This item is licensed under aCreative Commons License Creative Commons