Details regarding self-distillation

Hi,

Thanks for the great work! I have several questions regarding self-distillation:
#1. What kind of input sequence have you used, uniclust30_yyyy_mm_seed.fasta or uniclust30_yyyy_mm_consensus.fasta (I have fould two types of sequences in UniClust30), and what's the difference between them?

#2. There're uniclust30_XXX_a3m.ffdata and uniclust30_XXX_a3m_db in UniClust30. Have you used these MSAs?

#3. After generating and filtering MSAs and having 350k sequences, did you use bfd and pdb70 to generate related MSAs and templates and go through the inference phase to get the final self-distillation data?