导入PyTorch_Transformers时出现模块找不到错误

Question

通过Anaconda下载pytorch_transformers并通过Jupyter Notebook执行导入命令后，我遇到了几个与缺少模块相关的错误。

我尝试通过Anaconda搜索“sacremoses”以导入该包，但它仅适用于Linux机器。是否有其他人遇到过类似的问题？谢谢！

from pytorch_transformers import BertTokenizer、BertModel、BertForMaskedLM

这是错误信息:

---

<ipython-input-5-218d0858d00f> in <module>
----> 1 from pytorch_transformers import BertTokenizer, BertModel, BertForMaskedLM

~\Anaconda3\lib\site-packages\pytorch_transformers\__init__.py in <module>
      1 __version__ = "1.2.0"
----> 2 from .tokenization_auto import AutoTokenizer
      3 from .tokenization_bert import BertTokenizer, BasicTokenizer, WordpieceTokenizer
      4 from .tokenization_openai import OpenAIGPTTokenizer
      5 from .tokenization_transfo_xl import (TransfoXLTokenizer, TransfoXLCorpus)

~\Anaconda3\lib\site-packages\pytorch_transformers\tokenization_auto.py in <module>
     24 from .tokenization_transfo_xl import TransfoXLTokenizer
     25 from .tokenization_xlnet import XLNetTokenizer
---> 26 from .tokenization_xlm import XLMTokenizer
     27 from .tokenization_roberta import RobertaTokenizer
     28 from .tokenization_distilbert import DistilBertTokenizer

~\Anaconda3\lib\site-packages\pytorch_transformers\tokenization_xlm.py in <module>
     25 from io import open
     26 
---> 27 import sacremoses as sm
     28 
     29 from .tokenization_utils import PreTrainedTokenizer

ModuleNotFoundError: No module named 'sacremoses'```

- Jennifer Drake

Answer 1

请尝试按以下步骤创建conda环境并在创建的环境中安装软件包：

conda create -n env_pytorch -c intel python=3.5
source activate env_pytorch
pip install pytorch-transformers

- Lakshmi - Intel

---

谢谢您的回复，但是当我尝试运行'pip install pytorch-transformers'时，出现了以下错误： ERROR: Command errored out with exit status 1: python setup.py egg_info，请查看日志获取完整的命令输出。 - Jennifer Drake

Answer 2

你缺少pytorch-transformers的依赖。尝试使用以下命令安装它：

pip install pytorch-transformers

- Scott

Answer 3

尝试使用python -m pip install transformers而不是使用pip install来安装transformers库。

- Thanura Marapana