Adapting Large Language Models (LLMs) to Software Engineering (SE) tasks remains a major challenge. The main approach to achieve this is fine-tuning, which can be performed through resource-intensive methods such as Full Fine-Tuning (SFT) or lighter alternatives like Parameter-Efficient Fine-Tuning (PEFT). LoRA (Low-Rank Adaptation) and its enhanced version, QLoRA (Quantized Low-Rank Adaptation), are key examples of these efficient approaches. This paper reviews eight recent studies to examine how fine-tuning is applied in SE. The results reveal a clear shift from SFT to PEFT, particularly QLoRA, for tasks such as code repair and documentation generation. However, the reviewed studies use heterogeneous evaluation metrics, making cross-study comparison difficult. We conclude that while PEFT helps make LLMs more specialized and accessible, the field still requires standardized benchmarks and transparent efficiency reporting to support stronger and more comparable research.