在 spaCy 3.0 中加载管道时出错

如何解决在 spaCy 3.0 中加载管道时出错

更新到 spaCy 3.0.6 后，我无法加载任何经过训练的管道，尽管两者似乎都已正确安装：

================= Installed pipeline packages (spaCy v3.0.6) =================
ℹ spaCy installation:
/Users/baconbaker/anaconda3/envs/ml/lib/python3.8/site-packages/spacy

NAME              SPACY            VERSION                            
en_core_web_sm    >=3.0.0,<3.1.0   3.0.0   ✔
en_core_web_trf   >=3.0.0,<3.1.0   3.0.0   ✔

在使用 spacy.load() 并将管道作为模块导入时会发生这种情况（以下所有行的错误都相同）：

nlp = spacy.load("en_core_web_trf")

nlp = spacy.load("en_core_web_sm")

import en_core_web_sm
nlp = en_core_web_sm.load()

import en_core_web_trf
nlp = en_core_web_trf.load()

我得到的错误如下：

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-9-b38eb3aae320> in <module>
      1 import en_core_web_trf
----> 2 nlp = en_core_web_trf.load()

~/anaconda3/envs/ml/lib/python3.8/site-packages/en_core_web_trf/__init__.py in load(**overrides)
      8 
      9 def load(**overrides):
---> 10     return load_model_from_init_py(__file__,**overrides)

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy/util.py in load_model_from_init_py(init_file,vocab,disable,exclude,config)
    514     if not model_path.exists():
    515         raise IOError(Errors.E052.format(path=data_path))
--> 516     return load_model_from_path(
    517         data_path,518         vocab=vocab,~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy/util.py in load_model_from_path(model_path,Meta,config)
    389     config_path = model_path / "config.cfg"
    390     config = load_config(config_path,overrides=dict_to_dot(config))
--> 391     nlp = load_model_from_config(config,vocab=vocab,disable=disable,exclude=exclude)
    392     return nlp.from_disk(model_path,exclude=exclude)
    393 

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy/util.py in load_model_from_config(config,auto_fill,validate)
    426     # registry,including custom subclasses provided via entry points
    427     lang_cls = get_lang_class(nlp_config["lang"])
--> 428     nlp = lang_cls.from_config(
    429         config,430         vocab=vocab,~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy/language.py in from_config(cls,config,validate)
   1637         # then we would load them twice at runtime: once when we make from config,1638         # and then again when we load from disk.
-> 1639         nlp = lang_cls(vocab=vocab,create_tokenizer=create_tokenizer,Meta=Meta)
   1640         if after_creation is not None:
   1641             nlp = after_creation(nlp)

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy/language.py in __init__(self,max_length,create_tokenizer,batch_size,**kwargs)
    148         # points. The factory decorator applied to these functions takes care
    149         # of the rest.
--> 150         util.registry._entry_point_factories.get_all()
    151 
    152         self._config = DEFAULT_CONfig.merge(self.default_config)

~/anaconda3/envs/ml/lib/python3.8/site-packages/catalogue/__init__.py in get_all(self)
    106         result = {}
    107         if self.entry_points:
--> 108             result.update(self.get_entry_points())
    109         for keys,value in REGISTRY.items():
    110             if len(self.namespace) == len(keys) - 1 and all(

~/anaconda3/envs/ml/lib/python3.8/site-packages/catalogue/__init__.py in get_entry_points(self)
    121         result = {}
    122         for entry_point in AVAILABLE_ENTRY_POINTS.get(self.entry_point_namespace,[]):
--> 123             result[entry_point.name] = entry_point.load()
    124         return result
    125 

~/anaconda3/envs/ml/lib/python3.8/importlib/Metadata.py in load(self)
     75         """
     76         match = self.pattern.match(self.value)
---> 77         module = import_module(match.group('module'))
     78         attrs = filter(None,(match.group('attr') or '').split('.'))
     79         return functools.reduce(getattr,attrs,module)

~/anaconda3/envs/ml/lib/python3.8/importlib/__init__.py in import_module(name,package)
    125                 break
    126             level += 1
--> 127     return _bootstrap._gcd_import(name[level:],package,level)
    128 
    129 

~/anaconda3/envs/ml/lib/python3.8/importlib/_bootstrap.py in _gcd_import(name,level)

~/anaconda3/envs/ml/lib/python3.8/importlib/_bootstrap.py in _find_and_load(name,import_)

~/anaconda3/envs/ml/lib/python3.8/importlib/_bootstrap.py in _find_and_load_unlocked(name,import_)

~/anaconda3/envs/ml/lib/python3.8/importlib/_bootstrap.py in _call_with_frames_removed(f,*args,**kwds)

~/anaconda3/envs/ml/lib/python3.8/importlib/_bootstrap.py in _gcd_import(name,import_)

~/anaconda3/envs/ml/lib/python3.8/importlib/_bootstrap.py in _load_unlocked(spec)

~/anaconda3/envs/ml/lib/python3.8/importlib/_bootstrap_external.py in exec_module(self,module)

~/anaconda3/envs/ml/lib/python3.8/importlib/_bootstrap.py in _call_with_frames_removed(f,**kwds)

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy_transformers/__init__.py in <module>
----> 1 from . import architectures
      2 from . import annotation_setters
      3 from . import span_getters
      4 from .layers import TransformerModel
      5 from .pipeline_component import Transformer,install_extensions

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy_transformers/architectures.py in <module>
      3 from thinc.types import Ragged,Floats2d
      4 from spacy.tokens import Doc
----> 5 from .layers import TransformerModel,TransformerListener
      6 from .layers import trfs2arrays,split_trf_batch
      7 from .util import registry

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy_transformers/layers/__init__.py in <module>
----> 1 from .listener import TransformerListener
      2 from .transformer_model import TransformerModel
      3 from .split_trf import split_trf_batch
      4 from .trfs2arrays import trfs2arrays
      5 

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy_transformers/layers/listener.py in <module>
      2 from thinc.api import Model
      3 from spacy.tokens import Doc
----> 4 from ..data_classes import TransformerData
      5 
      6 

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy_transformers/data_classes.py in <module>
      9 import srsly
     10 
---> 11 from .util import transpose_list
     12 from .align import get_token_positions
     13 

~/anaconda3/envs/ml/lib/python3.8/site-packages/spacy_transformers/util.py in <module>
      2 from pathlib import Path
      3 import random
----> 4 from transformers import AutoModel,AutoTokenizer
      5 from transformers.tokenization_utils import BatchEncoding
      6 from transformers.tokenization_utils_fast import PreTrainedTokenizerFast

~/anaconda3/envs/ml/lib/python3.8/site-packages/transformers/__init__.py in <module>
    624 
    625     # Trainer
--> 626     from .trainer import Trainer
    627     from .trainer_pt_utils import torch_distributed_zero_first
    628 else:

~/anaconda3/envs/ml/lib/python3.8/site-packages/transformers/trainer.py in <module>
     67     TrainerState,68 )
---> 69 from .trainer_pt_utils import (
     70     distributedTensorGatherer,71     SequentialdistributedSampler,~/anaconda3/envs/ml/lib/python3.8/site-packages/transformers/trainer_pt_utils.py in <module>
     38     SAVE_STATE_WARNING = ""
     39 else:
---> 40     from torch.optim.lr_scheduler import SAVE_STATE_WARNING
     41 
     42 logger = logging.get_logger(__name__)

ImportError: cannot import name 'SAVE_STATE_WARNING' from 'torch.optim.lr_scheduler' (/Users/baconbaker/anaconda3/envs/ml/lib/python3.8/site-packages/torch/optim/lr_scheduler.py)

从当前稳定版本 1.8.1 恢复到 Torch 1.4.0 解决了问题，但我不想这样做。

有替代的解决方案吗？

解决方法

这似乎已在较新版本的 transformers (https://github.com/huggingface/transformers/pull/8979) 中得到修复。尝试同时升级 transformers 和 spacy-transformers。