I used Boltzgen with 100 num_designs and a budget of 10 and redesigned the best ranked sequences based on Boltzgen's ranking with ProteinMPNN with default parameters afterwards.