构建DeepPavlov的ner_ontonotes_bert_mult模型时遇TypeError错误求助

在Google Colab中按照DeepPavlov官方文档示例构建ner_ontonotes_bert_mult模型，使用代码：

from deeppavlov import build_model, configs
ner_model = build_model(configs.ner.ner_ontonotes_bert_mult, download=True) 

出现错误：TypeError: __init__() got an unexpected keyword argument 'num_tags'。补充说明：加载其他模型（如ner_rus_bert）时无此错误。

完整错误信息如下：

2022-12-17 13:08:23.235 INFO in 'deeppavlov.download'['download'] at line 138: Skipped http://files.deeppavlov.ai/v1/ner/ner_ontonotes_bert_mult_torch_crf.tar.gz download because of matching hashes
INFO:deeppavlov.download:Skipped http://files.deeppavlov.ai/v1/ner/ner_ontonotes_bert_mult_torch_crf.tar.gz download because of matching hashes
Some weights of the model checkpoint at bert-base-multilingual-cased were not used when initializing BertForTokenClassification: ['cls.predictions.transform.LayerNorm.weight', 'cls.predictions.transform.LayerNorm.bias', 'cls.predictions.bias', 'cls.predictions.decoder.weight', 'cls.seq_relationship.bias', 'cls.predictions.transform.dense.weight', 'cls.predictions.transform.dense.bias', 'cls.seq_relationship.weight']
- This IS expected if you are initializing BertForTokenClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing BertForTokenClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of BertForTokenClassification were not initialized from the model checkpoint at bert-base-multilingual-cased and are newly initialized: ['classifier.bias', 'classifier.weight']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
2022-12-17 13:08:30.1 ERROR in 'deeppavlov.core.common.params'['params'] at line 108: Exception in <class 'deeppavlov.models.torch_bert.torch_transformers_sequence_tagger.TorchTransformersSequenceTagger'>
Traceback (most recent call last):
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/core/common/params.py", line 102, in from_params
    component = obj(**dict(config_params, **kwargs))
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/torch_transformers_sequence_tagger.py", line 182, in __init__
    super().__init__(optimizer=optimizer,
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/core/models/torch_model.py", line 98, in __init__
    self.load()
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/torch_transformers_sequence_tagger.py", line 295, in load
    self.crf = CRF(self.n_classes).to(self.device)
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/crf.py", line 13, in __init__
    super().__init__(num_tags=num_tags, batch_first=batch_first)
TypeError: __init__() got an unexpected keyword argument 'num_tags'
ERROR:deeppavlov.core.common.params:Exception in <class 'deeppavlov.models.torch_bert.torch_transformers_sequence_tagger.TorchTransformersSequenceTagger'>
Traceback (most recent call last):
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/core/common/params.py", line 102, in from_params
    component = obj(**dict(config_params, **kwargs))
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/torch_transformers_sequence_tagger.py", line 182, in __init__
    super().__init__(optimizer=optimizer,
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/core/models/torch_model.py", line 98, in __init__
    self.load()
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/torch_transformers_sequence_tagger.py", line 295, in load
    self.crf = CRF(self.n_classes).to(self.device)
  File "/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/crf.py", line 13, in __init__
    super().__init__(num_tags=num_tags, batch_first=batch_first)
TypeError: __init__() got an unexpected keyword argument 'num_tags'
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-75-5e156706e7e4> in <module>
      1 # from deeppavlov import configs, build_model
      2 
----> 3 ner_model = build_model(configs.ner.ner_ontonotes_bert_mult, download=True)

5 frames
/usr/local/lib/python3.8/dist-packages/deeppavlov/core/commands/infer.py in build_model(config, mode, load_trained, install, download)
     53                             .format(component_config.get('class_name', component_config.get('ref', 'UNKNOWN'))))
     54 
---> 55         component = from_params(component_config, mode=mode)
     56 
     57         if 'id' in component_config:

/usr/local/lib/python3.8/dist-packages/deeppavlov/core/common/params.py in from_params(params, mode, **kwargs)
    100                 kwargs['mode'] = mode
    101 
---> 102             component = obj(**dict(config_params, **kwargs))
    103             try:
    104                 _refs[config_params['id']] = component

/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/torch_transformers_sequence_tagger.py in __init__(self, n_tags, pretrained_bert, bert_config_file, attention_probs_keep_prob, hidden_keep_prob, optimizer, optimizer_parameters, learning_rate_drop_patience, learning_rate_drop_div, load_before_drop, clip_norm, min_learning_rate, use_crf, **kwargs)
    180         self.use_crf = use_crf
    181 
---> 182         super().__init__(optimizer=optimizer,
    183                          optimizer_parameters=optimizer_parameters,
    184                          learning_rate_drop_patience=learning_rate_drop_patience,

/usr/local/lib/python3.8/dist-packages/deeppavlov/core/models/torch_model.py in __init__(self, device, optimizer, optimizer_parameters, lr_scheduler, lr_scheduler_parameters, learning_rate_drop_patience, learning_rate_drop_div, load_before_drop, min_learning_rate, *args, **kwargs)
     96         self.opt = deepcopy(kwargs)
     97 
---> 98         self.load()
     99         # we need to switch to eval mode here because by default it's in `train` mode.
    100         # But in case of `interact/build_model` usage, we need to have model in eval mode.

/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/torch_transformers_sequence_tagger.py in load(self, fname)
    293         self.model.to(self.device)
    294         if self.use_crf:
---> 295             self.crf = CRF(self.n_classes).to(self.device)
    296 
    297         self.optimizer = getattr(torch.optim, self.optimizer_name)(

/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/crf.py in __init__(self, num_tags, batch_first)
     11 
     12     def __init__(self, num_tags: int, batch_first: bool = False) -> None:
---> 13         super().__init__(num_tags=num_tags, batch_first=batch_first)
     14         nn.init.zeros_(self.transitions)
     15         nn.init.zeros_(self.start_transitions)

TypeError: __init__() got an unexpected keyword argument 'num_tags'

原因 #

错误源于DeepPavlov的CRF实现与当前环境中torchcrf库版本不兼容：

• DeepPavlov的crf.py中调用父类构造函数时使用了num_tags参数，但新版本torchcrf已将该参数改名为num_labels。
• ner_ontonotes_bert_mult模型依赖CRF组件，而ner_rus_bert未使用CRF，因此不受影响。

解决方法 #

方法1：降级torchcrf版本 #

安装兼容的旧版本torchcrf（0.7.2版本可解决该问题）：

pip install torchcrf==0.7.2

方法2：修改DeepPavlov的CRF代码 #

若不想降级库，可手动修改Colab中DeepPavlov的CRF实现文件：

1. 定位路径/usr/local/lib/python3.8/dist-packages/deeppavlov/models/torch_bert/crf.py
2. 将第13行的super().__init__(num_tags=num_tags, batch_first=batch_first)修改为：

super().__init__(num_labels=num_tags, batch_first=batch_first)

修改后重新运行模型加载代码即可。

内容的提问来源于stack exchange，提问作者Yuri