In order to alleviate server pressure, we are using the 650M version of the EsmFold model for structure prediction, while the results in our paper were evaluated by the 3B version. Please note that the prediction accuracy of the 650M version may be inferior compared to the 3B version. For further information, please refer to our publication: DIProT: A deep learning based interactive toolkit for efficient and effective Protein design. If you have any questions, feel free to contact us at he-jl23@mails.tsinghua.edu.cn