Small language models in higher education are specialized 3B- to 8B-parameter open-weight neural networks deployed local…