microsoft / DirectML

DirectML is a high-performance, hardware-accelerated DirectX 12 library for machine learning. DirectML provides GPU acceleration for common machine learning tasks across a broad range of supported hardware and drivers, including all DirectX 12-capable GPUs from vendors such as AMD, Intel, NVIDIA, and Qualcomm.
MIT License
2.19k stars 290 forks source link

OSError: [WinError 127] While trying to test Transformer sample #647

Open fabroGAMING opened 5 days ago

fabroGAMING commented 5 days ago

I wanted to check out how does DirectML work with transformers but when i tried following steps in attention_is_all_you_need the error occured.

Traceback (most recent call last): File "E:\Python_Projects\directml-test\DirectML\PyTorch\transformer\attention_is_all_you_need\train.py", line 40, in
from util.torchtext_legacy.field import Field
File "E:\Python_Projects\directml-test\DirectML\PyTorch\transformer\attention_is_all_you_need\util\torchtext_legacy\field.py", line 6, in from .data import Dataset File "E:\Python_Projects\directml-test\DirectML\PyTorch\transformer\attention_is_all_you_need\util\torchtext_legacy\data.py", line 13, in from torchtext.data.utils import RandomShuffler from torchtext import _extension # noqa: F401 File "E:\Python_Projects\directml-test\venv\lib\site-packages\torchtext_extension.py", line 64, in _init_extension() File "E:\Python_Projects\directml-test\venv\lib\site-packages\torchtext_extension.py", line 58, in _init_extension _load_lib("libtorchtext") File "E:\Python_Projects\directml-test\venv\lib\site-packages\torchtext_extension.py", line 50, in _load_lib torch.ops.load_library(path) File "E:\Python_Projects\directml-test\venv\lib\site-packages\torch_ops.py", line 1295, in load_library ctypes.CDLL(path) File "E:\Python\lib\ctypes__init.py", line 374, in init__ self._handle = _dlopen(self._name, mode) OSError: [WinError 127] Nie można odnaleźć określonej procedury

I tried reinstalling torch, directml and tried installing Windows SDK and Visual C++ packages but nothing worked.

fabroGAMING commented 5 days ago

Also i found that in requrements.txt there is a version of torchtext that downgrades torch that issnt supported by current DirectML

Edit: Whole example might be broken as now i think about it