"Bool value of tensor with more than one value is ambiguous" error when using "dp" but not "ddp"?

[EvanZ]

So I'm getting the following error which is not easy to track down:

"Bool value of tensor with more than one value is ambiguous"

It is throwing for the line of code that is doing the fitting:

trainer.fit(model, datamodule=dm)

This code has worked before on a cpu, but this is the first time I'm testing it on a GPU instance (a single 4-GPU instance running on EC2). It seems that this error occurs when I use 'dp' but runs without error when I switch to 'ddp'. Is this expected behavior?

[SkafteNicki]

It does not sound like an expected error.
If you could provide the full traceback we can better help you debug