Open niklaskorz opened 3 years ago
At least for allennlp 1.5.0, the fix seems to be quite easy:
Instead of accessing the former global DEFAULT_PREDICTORS
in my_allennlp/predictor/supertagger_predictor.py
, a property default_predictor
has to be defined in my_allennlp/models/supertagger.py
as part of the class:
@Model.register("supertagger")
class Supertagger(Model):
default_predictor = "supertagger-predictor"
...
I haven't tried whether this also works with 2.1.0 yet.
The current
my_allennlp
implementation tries to access variables that have been removed after version 0.9, e.g.DEFAULT_PREDICTORS
. The current version of AllenNLP is 2.1.0. I have tried installing 0.9.0 instead, but this seems to fail on my system with Python 3.8. I can install AllenNLP 1.5.0 and 2.1.0 just fine.