Patryk Ćwiek
Patryk Ćwiek
I encountered this ticket after more than a year with the exact same issue. We actually made it work. Not sure if it follows the 'best practices', but it's better...
Unfortunately not @jhedfactolerin I didn't have the chance to spend more time on it though. Ultimately I ended up post-processing the 'incorrectly' tokenized data into a more suitable shape.