Instructions for fine-tuning

[ardhe-qb]

Hi,

First of all – thanks much for all your work, what an amazing tool you've built here!

I am very curious to know if you had any form of formalized documentation on how one could fine-tune the model on a specific dataset? Interested mostly if there is anything to note when running training. I would also be interested in compute requirements, potential existing evaluation scripts, and so on.

Thanks in advance! Best,

Arnaud

[joewatchwell]

Training is described in some detail in the RFdiffusion supplement. That would be a good place to start :) In terms of evaluation, we evaluated on all of the benchmarks described in the paper, which should be pretty straightforward to set up

[Aminoacid1226]

Have you succeeded in fine-tuning? Can you provide some guidance?