Seeds vs diffusion samples

[GriffinYoung]

Can the authors give some guidance on whether it's better to generate different models using different seeds vs multiple diffusion samples?

[arogozhnikov]

re: previous answer: not that bad for AI, but code isn't syntactically correct.

We use both, that's a "default" way.

• feelin' lazy -> just run script once, get 5 samples
• want more - run with multiple seeds

For someone with a courage for experiments - there are many good question to poke around;

• when samples from the same trunk look alike?
• is it better to change a seed or reordering inputs in fasta?
• how many trunk recycles is 'good enough'? Does it depend on size of the complex?

Systematic study of such questions could form a very helpful paper.