Large Language Models (LLMs) such as Mistral, Mixtral, and Jamba are trained on large amounts of unstructured data and have significantly revolutionized the field of Natural Language Processing (NLP) through their ability to generate coherent and contextually relevant text. However, their performance in domain-specific tasks like medical question answering (QA), where precision is vital, still lags behind. This work investigates the potential of efficient instruction fine-tuning to enhance the performance of a large-scale, off-the-shelf LLM for biomedical language understanding. We introduce BioMed-LLaMa-3 8B, an instruction-tuned version of the 8-billion parameter Llama-3 model, trained on approximately 54K biomedical-focused instruction examples. Our model shows substantial improvements on the ChatDoctor dataset, with notable increases in BLEU, ROUGE, BERTScore, and METEOR metrics. These enhancements underscore the model’s potential for accurate medical QA. To foster further research, we make our code and models open-source ( https://github.com/zekaouinoureddine/BioMed-LLaMa-3 , https://huggingface.co/NouRed/BioMed-Tuned-Llama-3-8b ).