When source_text is a list of tokens make sure to join them with a space. We weren't doing this which caused the model to see citation text like: Krawczyketal.2013 which is not that kind of data the model was trained on.
I also join target_text with a space, but I havent seen a case where it is a list of length longer than 1.
When
source_textis a list of tokens make sure to join them with a space. We weren't doing this which caused the model to see citation text like:Krawczyketal.2013which is not that kind of data the model was trained on.I also join
target_textwith a space, but I havent seen a case where it is a list of length longer than 1.