First of all, thanks for this very clean implementation!
Upon running the biaffine dependency parser on Russian GSD treebank (using bert-base-multilingual-uncased), I've ran into a bug when preparing text for training:
ValueError: too many dimensions 'str'.
The problem seemed to be that the string version of UNK token was being used as encoding instead of its int version.
First of all, thanks for this very clean implementation!
Upon running the biaffine dependency parser on Russian GSD treebank (using bert-base-multilingual-uncased), I've ran into a bug when preparing text for training:
ValueError: too many dimensions 'str'
.The problem seemed to be that the string version of
UNK
token was being used as encoding instead of its int version.