In this thesis, we propose to fine-tune different pre-trained large language models to assist people in writing precise, concrete requirements. To achieve this goal, we choose two criteria(ISO- 29148 and Transformational Effects) to define a ’well-defined’ requirement. During training, we used different open-source 7 billion parameter LLMs(Zephyr-7b-beta, Llama2-7b-chat-hf, Gemma-1.1-7b-it) fine-tuned with the same customized dataset. With some auto-evaluation metrics(BertScore, Frugalscore, TER Score, BLEU Score, ROUGE Score, Exact Match) and manual evaluation, we draw the conclusion that Gemma-1.1-7b-it performs the best in our task. It has the potential to significantly accomplish our generating and rewriting tasks while detecting some of the transformational effects in the requirement.