ShadowProgrammer
ShadowProgrammer

Reputation: 82

"OSError: [WinError 127] The specified procedure could not be found" when importing torchtext

When importing get_tokenizer from torchtext:

from torchtext.data.utils import get_tokenizer

it throws this error:

Traceback (most recent call last):
  File "D:\PythonProjects\FakeNews\train.py", line 4, in <module>
    from torchtext.data.utils import get_tokenizer
  File "D:\PythonProjects\FakeNews\venv\Lib\site-packages\torchtext\__init__.py", line 18, in <module>
    from torchtext import _extension  # noqa: F401
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\PythonProjects\FakeNews\venv\Lib\site-packages\torchtext\_extension.py", line 64, in <module>
    _init_extension()
  File "D:\PythonProjects\FakeNews\venv\Lib\site-packages\torchtext\_extension.py", line 58, in _init_extension
    _load_lib("libtorchtext")
  File "D:\PythonProjects\FakeNews\venv\Lib\site-packages\torchtext\_extension.py", line 50, in _load_lib
    torch.ops.load_library(path)
  File "D:\PythonProjects\FakeNews\venv\Lib\site-packages\torch\_ops.py", line 1295, in load_library
    ctypes.CDLL(path)
  File "C:\Users\Owner\AppData\Local\Programs\Python\Python312\Lib\ctypes\__init__.py", line 379, in __init__
    self._handle = _dlopen(self._name, mode)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^
OSError: [WinError 127] The specified procedure could not be found

I'm using torch 2.4.0 and torchtext 0.18.0, which should be compatible.

Reinstalling and --upgrade did not work. Rerunning just throws the same error. No forums I've found say anything helpful either.

Upvotes: 1

Views: 1811

Answers (2)

anon ymous
anon ymous

Reputation: 1

No forums I've found say anything helpful either.

I have the exact same feeling. No forums are helping with this error.

Do you have an error window pop up (if you are using Windows) that says this?

Error window

EDIT: I've solved the problem for myself. I just needed to downgrade to torch-2.2.2 and torchtext-0.17.2. To do this:

pip install torch==2.2.2 torchtext==0.17.2

Upvotes: 0

ShadowProgrammer
ShadowProgrammer

Reputation: 82

Turns out I was using PyTorch for CUDA 12.4 but torchtext only supports CUDA 12.1, so I had to change the compute platform.

TLDR; torchtext 0.18.0 and pytorch 2.4.0 work together on CUDA 12.1 but not 12.4

Upvotes: 0

Related Questions