Joachim Wagner

However, the `config.json` of a downloaded model suggests that the model was not trained on a conllu file: `"train_path": "/users4/conll18st/raw_text/Czech/cs-20m.raw"`. Has this historic reasons, i.e. was conllu input format only...

Related: issue #402

It works if you move the line `` up above the line `` in the intermediate html. (I should also mention that I removed the redundant `` in ``, in...

> [...] If you think this still needs to be addressed please comment on this thread. This feature would have many applications and would enable comparison of MLMs in gloze...

> @jowagner Just to clarify it for others who might be following, the paper you are referring to is this one https://arxiv.org/abs/2002.03079 right? Yes. I hope to read it soon...

Looking at Devlin et al 2018 again, I don't see the pre-training objective stated but certainly they try to push as much probability mass as possible to the one completion...

> This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread....