Thanks for the report. This is strange and definitely not intended and one set shouldn’t contain examples for the same hashes. Did you spot the duplicate examples towards the end of the dataset or last? This would indicate that maybe there’s an unintended overlap somewhere.
I suspect that the cause of this problem might be related to this issue: Old examples are automatically added to new dataset. I’m pretty sure we tracked this down to a problem with how the database handles the input and task hashes, and we’ve already fixed that for the upcoming v1.4.1 (just running a few test and will release the update asap, so you can check if it fixes the problem).