Gradient Checkpointing Error

[shenoynikhil]

I was trying to use gradient checkpointing with TorchMD model. For some reason, I get this gradient mismatch whenever the warning pops up

torchmd-net/torchmdnet/models/utils.py

Lines 492 to 499 in 8a1be71

	if not mask.all():
	warnings.warn(
	(
	f"Skipping gradients for {(~mask).sum()} atoms due to vector features being zero. "
	"This is likely due to atoms being outside the cutoff radius of any other atom. "
	"These atoms will not interact with any other atom unless you change the cutoff."
	)
	)

[RaulPPelaez]

I believe this should be ok to ignore if you are ok with its message (a.i. you know the cutoff is such that some atoms are left alone). If you know for a fact this is not the case but the message pops up anyway then this is a bug.