Lightweight libraries that read text streams directly, such as PyPDF and pdfplumber, do not load neural network models and typically consume memory on the order of tens of MB.