question about loss function

[xtanh]

Hi, great work! I noticed that although the paper mentions a KL divergence term in the loss (Eq. 12), it doesn’t seem to be implemented in the training code—was it actually used in practice?

[kamzero]

Thanks for your interest! The KL divergence term in Equation 12 was omitted from our open-source training code, as we discovered post-submission that omitting it still achieves strong performance on ddG prediction task. However, it remains useful for preserving the inverse folding model's original capabilities.