Custom>TFBertMainLayer已注册到下载模型时发生的<class'src.transformers.modeling\u tf\u bert.TFBertMainLayer'>

Question

我一直在试着玩预先训练好的模特儿。在协作模板中，所有内容都处于默认状态，使用torch.hub.load()从huggingface/pytorch-transformers到bert-base-uncased作为“模型”

代码示例

import torch
model = torch.hub.load('huggingface/pytorch-transformers', 'model', 'bert-base-uncased')    # Download model and configuration from S3 and cache.

我遇到了这个错误

Using cache found in /root/.cache/torch/hub/huggingface_pytorch-transformers_master
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-19-ad22a1a34951> in <module>()
      1 import torch
----> 2 model = torch.hub.load('huggingface/pytorch-transformers', 'model', 'somethingelse')    # Download model and configuration from S3 and cache.
      3 model = torch.hub.load('huggingface/pytorch-transformers', 'model', './test/bert_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
      4 model = torch.hub.load('huggingface/pytorch-transformers', 'model', 'bert-base-uncased', output_attentions=True)  # Update configuration during loading
      5 assert model.config.output_attentions == True

13 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/utils/generic_utils.py in decorator(arg)
    177       raise ValueError(
    178           '%s has already been registered to %s' %
--> 179           (registered_name, _GLOBAL_CUSTOM_OBJECTS[registered_name]))
    180 
    181     if arg in _GLOBAL_CUSTOM_NAMES:

ValueError: Custom>TFBertMainLayer has already been registered to <class 'src.transformers.modeling_tf_bert.TFBertMainLayer'>

我真的不明白发生了什么

Answer 1

如果您只想使用PyTorch版本的transformer，为了解决这个问题，您可以从您的环境中卸载TensorFlow，或者在transformer的源代码中注释掉init.py中的TensorFlow部分：）