Current issue for the TabNet learners:

[train()] learner 'classif.torch.tabnet:response' on task 'feat_all_binary' failed: 
train log has errors: R6 class dataloader/R6 does not have slot '.IterableDataset_len_called'!

This seems to be a {torch} thing, and I have not identified the root cause yet.
In addition, I'm not sure if the fix would be to adjust the autotest (PR to mlr3?) for torch-specifics or to adjust the learner, but that'll depend on the actual cause of the issue.