Hi, i'm actually using the rel.manual module to create a relation between different entities extracted from a custom NER model, i used different dataset (5 in total) with different text in each, and every time i get this strange error after labelling the 3 rd text of each dataset (maybe this is a clue) :
i'm using prodigy 1.11.7
Edit : for more context, it's happening only when i use the "wrap" feature of this recipe, which is absolutely required for my texts to be readable easily
Thanks for the report! Can you share an example of the data you're loading in? Is it normal raw data or does it contain already annotated spans etc.?
It is normal raw data, from a json file not a jsonl, the text is not already annotated, but i used a pretrained model that perform NER on it at the same time that i annotate relation between entities found
i can not share an example because it is sensible data sorry, but i can tel you that it's a classic text, with no pre-processing apply to it, approximately 500-1000 token
That's definitely strange 
Totally understand about not being able to share the data, but can you reproduce this by using similar text where you replace the actual words with some gibberish? And is there anything special about the texts, like whitespace tokens or similar?
