<p>In the pursuit of efficient automated text summarization, several language models have been developed which include sophisticated transformer language models like T5, PEGASUS by Google, and BART by Facebook. In this study, we aim to perform a comprehensive evaluation of the performance of two large-language models: FLAN T5 and LLaMA 2 fine-tuned using the parameter-efficient fine-tuning (PEFT) technique Low-Rank Adaptation, on benchmark datasets - SAMSum and DialogSum. Additionally, we also test the efficacy of the models on a custom test set. The key metrics that serve as the basis for our evaluation include ROUGE scores, BLEU scores, Latent Semantic Analysis similarity scores, and BERT scores. The large language models FLAN-T5 and LLaMA 2 were fine-tuned using parameter-efficient fine-tuning techniques called Low-Rank Adaptation (LoRA) and Quantized Low-Rank Adaptation (QLoRA). The evaluation metrics employed were ROUGE (for n-gram overlap), BERTScore (for semantic similarity), BLEU (for translation quality), and LSA (for latent semantic analysis). LLaMA 2 showed excellent performance with the highest ROUGE-1 score of 48.97%, LSA of 80.79%, BERT score of 91% on the SAMSum dataset, and ROUGE-1 score of 48.60%, LSA of 74.95%, 92% BERT score on DialogSum dataset. The same trend was observed on the custom test set with the LLaMA 2 model fine-tuned on the DialogSum dataset achieving a ROUGE-1 score of 44% reiterating the efficiency of PEFT using QLoRA an enhanced method of LoRA that introduces quantization to efficiently handle the computation of parameters during fine-tuning. The results underscore the potential of large language models and highlight the importance of further research in memory-optimized techniques for fine-tuning. Overcoming the challenges of computation cost and hardware requirements in full fine-tuning, this study highlights the versatility of large-language models, positioning them as a promising foundation for advanced natural language processing tasks in the future.</p>