LayoutLMV3 inference without label

how to test images using LayoutLMV3 model without labels/annotations?