Saving and loading a pipeline with a coreferee component

[riccardopinosio]

Hello,

I'm trying to use coreferee with spacy. If I do:

import spacy
import coreferee

nlp = spacy.load("en_core_web_lg")
nlp.add_pipe("coreferee")
nlp.to_disk("./test_pipe")
nlp = spacy.load("./test_pipe")

I get:

KeyError                                  Traceback (most recent call last)
File ~/miniconda3/envs/relationlearner/lib/python3.10/site-packages/spacy/language.py:1020, in Language.__call__(self, text, disable, component_cfg)
   1019 try:
-> 1020     doc = proc(doc, **component_cfg.get(name, {}))  # type: ignore[call-arg]
   1021 except KeyError as e:
   1022     # This typically happens if a component is not initialized

File ~/miniconda3/envs/relationlearner/lib/python3.10/site-packages/spacy/pipeline/trainable_pipe.pyx:56, in spacy.pipeline.trainable_pipe.TrainablePipe.__call__()

File ~/miniconda3/envs/relationlearner/lib/python3.10/site-packages/spacy/util.py:1658, in raise_error(proc_name, proc, docs, e)
   1657 def raise_error(proc_name, proc, docs, e):
-> 1658     raise e

File ~/miniconda3/envs/relationlearner/lib/python3.10/site-packages/spacy/pipeline/trainable_pipe.pyx:52, in spacy.pipeline.trainable_pipe.TrainablePipe.__call__()

File ~/miniconda3/envs/relationlearner/lib/python3.10/site-packages/spacy/pipeline/tok2vec.py:125, in Tok2Vec.predict(self, docs)
    124     return [self.model.ops.alloc((0, width)) for doc in docs]
--> 125 tokvecs = self.model.predict(docs)
    126 batch_id = Tok2VecListener.get_batch_id(docs)

File ~/miniconda3/envs/relationlearner/lib/python3.10/site-packages/thinc/model.py:315, in Model.predict(self, X)
    312 """Call the model's `forward` function with `is_train=False`, and return
    313 only the output, instead of the `(output, callback)` tuple.
    314 """
...
-> 1023     raise ValueError(Errors.E109.format(name=name)) from e
   1024 except Exception as e:
   1025     error_handler(name, proc, [doc], e)

ValueError: [E109] Component 'tok2vec' could not be run. Did you forget to call `initialize()`?

I'm not sure what's happening here. In my understanding coreferee should not mess with the tok2vec component of the pipeline. It also doesn't seem to matter where coreferee is added in the pipeline, it will always raise that error.
Any ideas?

[adrianeboyd]

This looks like a bug. There's an in-progress PR at: richardpaulhudson/coreferee#14

It looks like to_disk/from_disk will currently work with en_core_web_trf, but not en_core_web_sm/md/lg. I can't think of a workaround, so you may need to wait until the next coreferee release for this to be fixed.