The more I google the more posts I find about this. Understanding code - error Expected more than 1 value per channel when training
This does trip people up every once in a while. It can be especially annoying with the new fastai v1 library since the size of the train and validation set varies between calls to random_split which makes this error unpredictable.
Jeremy I am happy to implement whatever solution you think is best, but I believe this does require a design choice. Shall we throw an error explaining what is going on and let the user handle it? Or shall we silently remove the offending batch?