Log the same metric when using multiple dataloaders in validation_step()

[thangld201]

Hi, I have a particular use case when using multiple dataloaders for validation in validation_step(). I have metrics that I want to aggregate across these different dataloaders for checkpoint monitoring, but when I use self.log('val/loss',....) in the validation_step(), automatically a suffix will be added: e.g. 'val/loss' -> 'val/loss/dataloader_idx_0', 'val/loss' -> 'val/loss/dataloader_idx_1'... This makes it impossible to reduce this metric across datasets.

I tried setting add_dataloader_idx=False but it reports another error

pytorch_lightning.utilities.exceptions.MisconfigurationException: You called 
`self.log(val/loss, ...)` twice in `validation_step` with different arguments. This is not allowed

How can I do this in Pytorch Lightning ? I'm currently using the 1.9.4 version.

[SeppeLampe]

It's been a while, but did you find a solution?