A Comprehensive Review of LLMs: Architecture, Performance, and Applications
12th International Conference on Electrical, Electronic and Computing Engineering, IcETRAN 2025, Cacak, Sırbistan, 9 - 12 Haziran 2025, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/icetran66854.2025.11114289
- Basıldığı Şehir: Cacak
- Basıldığı Ülke: Sırbistan
- Anahtar Kelimeler: context modeling, large language models (LLMs), software engineering, transformers
- Süleyman Demirel Üniversitesi Adresli: Evet
Özet
This study presents a comprehensive review of LLMs, which have emerged with the developments in the fields of AI and NLP in recent years and are built on deep learning architectures. LLMs, which have created a revolutionary impact with their recent use in various fields such as code generation, error detection, text summarization, and multilingual support, are actively used in many fields such as healthcare, finance, chatbots, autonomous systems, and academic research. LLMs are divided into three main architectures: decoder, encoder, and decoder-encoder models. Within the scope of the study, a comprehensive comparison of the features of popular methods in these three main architectures has been made. The common features of these methods have been tested in detail under equal conditions. The obtained findings indicate that CodeT5 is the fastest with a runtime of less than a second, while CodeGen outperforms the alternatives with a BertScore value of 0.839.