Loading...
Publication
Otimização de prompts em Modelos de Linguagem de Grande Escala
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.73 MB | Adobe PDF |
Advisor(s)
Abstract(s)
A presente dissertação aborda o impacto das técnicas de prompt engineering na utilização de Modelos de Linguagem de Grande Escala (LLMs), com foco no domínio da programação. Apesar do enorme potencial destes modelos, a sua eficácia depende fortemente da formulação de instruções claras, contextualizadas e adequadas. Neste trabalho é proposto o desenvolvimento de um sistema capaz de gerar versões otimizadas de prompts fornecidos por utilizadores, aplicando estratégias de melhoria contínua e feedback. O sistema é avaliado através da comparação entre as respostas produzidas a partir dos prompts originais e otimizados, com recurso a métricas como precisão, e execução prática do código gerado. A investigação contribui para o aprofundamento do conhecimento sobre a engenharia de prompts e para a criação de interações mais eficazes e fiáveis com LLMs, oferecendo suporte prático a programadores e profissionais que dependem da qualidade das respostas para
resolver problemas técnicos.
This dissertation investigates the role of prompt engineering techniques in improving the performance of Large Language Models (LLMs) within the programming domain. While LLMs have shown remarkable potential across multiple areas, their practical effectiveness is often limited by the quality and contextual relevance of the prompts provided. To address this challenge, we propose the development of a system designed to generate optimized versions of user-provided prompts, making use of prompt engineering strategies such as clarification, chain-of-thought reasoning, and task decomposition. The system was implemented in Python with widely adopted libraries and frameworks, including OpenAI’s API and Gradio for the user interface. The evaluation process focused on the comparative analysis of outputs generated with original prompts versus optimized prompts, employing metrics such as accuracy and the executable validity of the produced code. Overall, this work provides both theoretical and practical insights into prompt engineering, aiming to support researchers, developers, and practitioners in building more effective interactions with LLMs.
This dissertation investigates the role of prompt engineering techniques in improving the performance of Large Language Models (LLMs) within the programming domain. While LLMs have shown remarkable potential across multiple areas, their practical effectiveness is often limited by the quality and contextual relevance of the prompts provided. To address this challenge, we propose the development of a system designed to generate optimized versions of user-provided prompts, making use of prompt engineering strategies such as clarification, chain-of-thought reasoning, and task decomposition. The system was implemented in Python with widely adopted libraries and frameworks, including OpenAI’s API and Gradio for the user interface. The evaluation process focused on the comparative analysis of outputs generated with original prompts versus optimized prompts, employing metrics such as accuracy and the executable validity of the produced code. Overall, this work provides both theoretical and practical insights into prompt engineering, aiming to support researchers, developers, and practitioners in building more effective interactions with LLMs.
Description
Keywords
Artificial Intelligence Large Language Models Prompt Engineering Inteligência artificial Modelos de Linguagem em Grande Escala Engenharia de prompts