Adapting Large Language Models (LLMs) for a downstream task is a multi-phase process that involves understanding the new knowledge domain and fine-tuning the model for the given task. This process requires sizable budgets due to the large number of parameters needed to acquire and capture representative semantics. In recent years, there have been breakthroughs in training LLMs with parameter-efficient techniques, making it possible for LLMs to be fine-tuned for a desired task by requiring only a fraction of the optimizable parameters. In this work, we employ the Llama-2 model by Meta with 7B parameters and fine-tune it for the Question-Answering task in a new knowledge domain. Two different datasets are used, including a collection of physics books and a recent paper. The two models successfully learned the new knowledge domain, responding to queries with correct, representative answers from the text corpus. Additionally, samples of questions and generated answers are provided.