error in add_pipe()

[bisserai]

hello community,

i'm very unexperienced, so apologies if there is an obvious solution i've missed. i'm trying to use benepar to get the constituency structure of some german texts. when i go
nlp = spacy.load('de_core_news_lg') nlp.add_pipe("benepar", config={"model": "benepar_de2"})
i unfortunately get the error:
RuntimeError: Error(s) in loading state_dict for ChartParser:
Unexpected key(s) in state_dict: "pretrained_model.embeddings.position_ids".

i think there might also be a problem with my environment as in the import statements i get the following error:
/Users/ivanova/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/torch/nn/modules/transformer.py:20: UserWarning: Failed to initialize NumPy: _ARRAY_API not found (Triggered internally at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/utils/tensor_numpy.cpp:84.)
device: torch.device = torch.device(torch._C._get_default_device()), # torch.device('cpu'),
i tried installing different versions of numpy, but it didnt' help.

thanks for your help,
bissera

full traceback of main error:

RuntimeError Traceback (most recent call last)
Cell In[4], line 2
1 nlp = spacy.load('de_core_news_lg')
----> 2 nlp.add_pipe("benepar", config={"model": "benepar_de2"})
3 # if spacy.version.startswith('2'):
4 # nlp.add_pipe(benepar.BeneparComponent("benepar_de2"))
5 # else:
6 # nlp.add_pipe("benepar", config={"model": "benepar_de2"})

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/spacy/language.py:824, in Language.add_pipe(self, factory_name, name, before, after, first, last, source, config, raw_config, validate)
820 pipe_component, factory_name = self.create_pipe_from_source(
821 factory_name, source, name=name
822 )
823 else:
--> 824 pipe_component = self.create_pipe(
825 factory_name,
826 name=name,
827 config=config,
828 raw_config=raw_config,
829 validate=validate,
830 )
831 pipe_index = self._get_pipe_index(before, after, first, last)
832 self._pipe_meta[name] = self.get_factory_meta(factory_name)

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/spacy/language.py:712, in Language.create_pipe(self, factory_name, name, config, raw_config, validate)
709 cfg = {factory_name: config}
710 # We're calling the internal _fill here to avoid constructing the
711 # registered functions twice
--> 712 resolved = registry.resolve(cfg, validate=validate)
713 filled = registry.fill({"cfg": cfg[factory_name]}, validate=validate)["cfg"]
714 filled = Config(filled)

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/confection/init.py:760, in registry.resolve(cls, config, schema, overrides, validate)
751 @classmethod
752 def resolve(
753 cls,
(...)
758 validate: bool = True,
759 ) -> Dict[str, Any]:
--> 760 resolved, _ = cls._make(
761 config, schema=schema, overrides=overrides, validate=validate, resolve=True
762 )
763 return resolved

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/confection/init.py:809, in registry._make(cls, config, schema, overrides, resolve, validate)
807 if not is_interpolated:
808 config = Config(orig_config).interpolate()
--> 809 filled, _, resolved = cls._fill(
810 config, schema, validate=validate, overrides=overrides, resolve=resolve
811 )
812 filled = Config(filled, section_order=section_order)
813 # Check that overrides didn't include invalid properties not in config

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/confection/init.py:881, in registry._fill(cls, config, schema, validate, resolve, parent, overrides)
878 getter = cls.get(reg_name, func_name)
879 # We don't want to try/except this and raise our own error
880 # here, because we want the traceback if the function fails.
--> 881 getter_result = getter(*args, **kwargs)
882 else:
883 # We're not resolving and calling the function, so replace
884 # the getter_result with a Promise class
885 getter_result = Promise(
886 registry=reg_name, name=func_name, args=args, kwargs=kwargs
887 )

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/benepar/integrations/spacy_plugin.py:176, in create_benepar_component(nlp, name, model, subbatch_max_tokens, disable_tagger)
169 def create_benepar_component(
170 nlp,
171 name,
(...)
174 disable_tagger: bool,
175 ):
--> 176 return BeneparComponent(
177 model,
178 subbatch_max_tokens=subbatch_max_tokens,
179 disable_tagger=disable_tagger,
180 )

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/benepar/integrations/spacy_plugin.py:116, in BeneparComponent.init(self, name, subbatch_max_tokens, disable_tagger, batch_size)
96 def init(
97 self,
98 name,
(...)
101 batch_size="ignored",
102 ):
103 """Load a trained parser model.
104
105 Args:
(...)
114 batch_size: deprecated and ignored; use subbatch_max_tokens instead
115 """
--> 116 self._parser = load_trained_model(name)
117 if torch.cuda.is_available():
118 self._parser.cuda()

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/benepar/integrations/downloader.py:34, in load_trained_model(model_name_or_path)
32 model_path = locate_model(model_name_or_path)
33 from ..parse_chart import ChartParser
---> 34 parser = ChartParser.from_trained(model_path)
35 return parser

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/benepar/parse_chart.py:186, in ChartParser.from_trained(cls, model_path)
184 config["hparams"] = nkutil.HParams(**hparams)
185 parser = cls(**config)
--> 186 parser.load_state_dict(state_dict)
187 return parser

File ~/opt/anaconda3/envs/parseg/lib/python3.12/site-packages/torch/nn/modules/module.py:2153, in Module.load_state_dict(self, state_dict, strict, assign)
2148 error_msgs.insert(
2149 0, 'Missing key(s) in state_dict: {}. '.format(
2150 ', '.join(f'"{k}"' for k in missing_keys)))
2152 if len(error_msgs) > 0:
-> 2153 raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
2154 self.class.name, "\n\t".join(error_msgs)))
2155 return _IncompatibleKeys(missing_keys, unexpected_keys)

RuntimeError: Error(s) in loading state_dict for ChartParser:
Unexpected key(s) in state_dict: "pretrained_model.embeddings.position_ids".

[bisserai]

update: i get exactly the same error when i try to run with nltk instead