Add Support for Single-Sentence Protein Prediction Model (e.g. ESM2 or ESMFold) ?

[pengshuang]

Thanks for releasing the OpenFold code for structure prediction！

However, I wonder if it is possible to train the ESMFold model based on OpenFold. To be specific, I want to finetune the protein language model ESM-2 on the large-scale protein sequence database (e.g. UniRef 90). And I can get a new model for the downstream protein folding task. I notice that the ESMFold code partly relies on openfold, and I want to know whether is it possible to finish the training process of the ESMFold model based on the openfold code.

Hope for your suggestions. Thanks in advance!