The number of training parameters and the computational costs have grown exponentially with the increased efficiency of large language models. These large language models (LLMs) have remarkable linguistic and generative abilities that can be leveraged using fine-tuning to optimize their performance across specific tasks. Fine-tuning involves training with fewer parameters, making it more effective over certain domains, and it also utilizes various approaches to cater to varied outputs as required by the particular task. Fine-tuning is a significant technique in the context of current advancements in NLP, with contributions like multilingual chatbots, personalized healthcare assistants, combating misinformation, and shaping the future with few-shot learning and multimodal fine-tuning, thus improving their performance and applicability in real-world scenarios. Reviews of fine-tuning strategies for large language models offer insights into challenge mitigation, prompt engineering, regularization methods, computational efficiency, and techniques like retrieval-augmented generation (RAG), enhancing the performance and applicability of these models. This paper comprehensively dives deep into eight prominent fine-tuning strategies, including instruction tuning, transfer learning, basic hyperparameter tuning, retrieval-augmented generation (RAG), parameter-efficient tuning (PEFT), reward modeling, preference learning, and proximal policy optimization (PPO), which significantly enhances the efficiency, scalability, and interpretability of LLMs, followed by a conclusion and future prospectus. Each approach has been reviewed for its unique advantages, challenges, and working methodology for fine-tuning the LLMs.