This work addresses the need for a balanced approach between performance and efficiency
in scalable production environments for visually-rich document understanding (VDU) tasks …