Sort models by Q4 memory requirement before running compatibility checks. This avoids downloading models that are outside your GPU memory envelope.