python3.12使用paddlehub出现报错Error loading \torch\lib\shm.dll

Question

问题

python3.12使用paddlehub出现报错

代码如下

import os
import cv2
import paddlehub as hub
# 显式指定使用 GPU
# os.environ['PADDLE_INFER_DEVICE'] = 'gpu'
ocr = hub.Module(name='chinese_ocr_db_crnn_mobile')
image_path = 'img.png'
img = cv2.imread(image_path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
results = ocr.recognize_text(images=[img], use_gpu=False)
for result in results:
    print(result['data'])

报错如下

D:\pythonobj\ocrtext\.venv\Lib\site-packages\jieba\_compat.py:18: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
  import pkg_resources
Traceback (most recent call last):
  File "D:\pythonobj\ocrtext\text6.py", line 3, in <module>
    import paddlehub as hub
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlehub\__init__.py", line 31, in <module>
    from paddlehub import datasets
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlehub\datasets\__init__.py", line 16, in <module>
    from paddlehub.datasets.chnsenticorp import ChnSentiCorp
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlehub\datasets\chnsenticorp.py", line 19, in <module>
    from paddlehub.datasets.base_nlp_dataset import TextClassificationDataset
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlehub\datasets\base_nlp_dataset.py", line 21, in <module>
    import paddlenlp
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\__init__.py", line 35, in <module>
    from . import (
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\dataaug\__init__.py", line 17, in <module>
    from .sentence import *
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\dataaug\sentence.py", line 17, in <module>
    from ..taskflow import Taskflow
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\taskflow\__init__.py", line 15, in <module>
    from .taskflow import Taskflow
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\taskflow\taskflow.py", line 26, in <module>
    from .information_extraction import GPTask, UIETask
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\taskflow\information_extraction.py", line 26, in <module>
    from ..datasets import load_dataset
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\datasets\__init__.py", line 15, in <module>
    from .bellegroup import *
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\datasets\bellegroup.py", line 22, in <module>
    from .dataset import DatasetBuilder
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\paddlenlp\datasets\dataset.py", line 23, in <module>
    import datasets
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\datasets\__init__.py", line 17, in <module>
    from .arrow_dataset import Column, Dataset
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\datasets\arrow_dataset.py", line 104, in <module>
    from .formatting import format_table, get_format_type_from_alias, get_formatter, query_table
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\datasets\formatting\__init__.py", line 91, in <module>
    from .torch_formatter import TorchFormatter
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\datasets\formatting\torch_formatter.py", line 32, in <module>
    import torch
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\torch\__init__.py", line 281, in <module>
    _load_dll_libraries()
  File "D:\pythonobj\ocrtext\.venv\Lib\site-packages\torch\__init__.py", line 264, in _load_dll_libraries
    raise err
OSError: [WinError 127] 找不到指定的程序。 Error loading "D:\pythonobj\ocrtext\.venv\Lib\site-packages\torch\lib\shm.dll" or one of its dependencies.

环境

CUDA Version: 12.7

paddleocr 3.3.0

python -m pip install paddlepaddle-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu129/

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu129

paddlehub 2.4.0

Answer 1

找到原因：paddlehub和paddleOCR的包使用的pytorch指定的包内容，和官方的包区别较大。需要和其他使用pytorch的包区别开来。