Diffusion inferers currently do not support classifier-free guidance; the classifier-free guidance tutorial (https://github.com/Project-MONAI/tutorials/tree/main/generation/classifier_free_guidance) has to implement the sample method outside of the inferer.
Describe the solution you'd like
Perhaps, there could be a classifier-free guidance attribute to the inferer that branches the sample logic so that, if not None, the sampling method samples an unconditioned / conditioned output for every t in the for loop as in the tutorial. This would not alter the behaviour if the attribute is defaulted to None.
Diffusion inferers currently do not support classifier-free guidance; the classifier-free guidance tutorial (https://github.com/Project-MONAI/tutorials/tree/main/generation/classifier_free_guidance) has to implement the sample method outside of the inferer.
Describe the solution you'd like
Perhaps, there could be a classifier-free guidance attribute to the inferer that branches the sample logic so that, if not None, the sampling method samples an unconditioned / conditioned output for every t in the for loop as in the tutorial. This would not alter the behaviour if the attribute is defaulted to None.