Based on the sampling results of multiple models in the paper and my own sampling results, I am confused by the model's unconditional sampling preference for helix structures. Especially during the training process, the models at different steps showed great differences. Some models preferred α-helix, while others preferred β-sheet. Do you have any idea about this issue?
Based on the sampling results of multiple models in the paper and my own sampling results, I am confused by the model's unconditional sampling preference for helix structures. Especially during the training process, the models at different steps showed great differences. Some models preferred α-helix, while others preferred β-sheet. Do you have any idea about this issue?