ImportError: cannot import name 'DisableTorchFunctionSubclass' from 'torch._C'

[lukazso]

🐛 Describe the bug

Hi all,

I was recently trying to install torchvision from the main branch to include the prototype features. I followed the install procedure specified on the contributor page:

pip install numpy
pip install --pre torch -f https://download.pytorch.org/whl/nightly/cu117/torch_nightly.html

git clone https://github.com/pytorch/vision.git
cd vision
python setup.py develop
pip install flake8 typing mypy pytest pytest-mock scipy

However, when I try to import the protoype features, I run into following problem:

>>> from torchvision.prototype import datapoints, transforms as T
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/vision/torchvision/prototype/__init__.py", line 1, in <module>
    from . import datapoints, models, transforms, utils
  File "/vision/torchvision/prototype/datapoints/__init__.py", line 1, in <module>
    from ._bounding_box import BoundingBox, BoundingBoxFormat
  File "/vision/torchvision/prototype/datapoints/_bounding_box.py", line 9, in <module>
    from ._datapoint import Datapoint, FillTypeJIT
  File "/vision/torchvision/prototype/datapoints/_datapoint.py", line 8, in <module>
    from torch._C import DisableTorchFunctionSubclass
ImportError: cannot import name 'DisableTorchFunctionSubclass' from 'torch._C' (/usr/local/lib/python3.8/dist-packages/torch/_C.cpython-38-x86_64-linux-gnu.so)

I tried it in a virtual environment as well as a Docker container, both the same results.

Versions

Environment in Docker:

PyTorch version: 1.13.1+cu117
Is debug build: False
CUDA used to build PyTorch: 11.7
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.5 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.31

Python version: 3.8.10 (default, Nov 14 2022, 12:59:47)  [GCC 9.4.0] (64-bit runtime)
Python platform: Linux-5.15.0-50-generic-x86_64-with-glibc2.29
Is CUDA available: False
CUDA runtime version: 11.7.99
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: Could not collect
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

Versions of relevant libraries:
[pip3] mypy==0.991
[pip3] mypy-extensions==0.4.3
[pip3] numpy==1.24.1
[pip3] torch==1.13.1
[pip3] torchvision==0.15.0a0+5dd9594
[conda] Could not collect

Environment on local machine in virtual environment:

Collecting environment information...
PyTorch version: 1.13.1+cu117
Is debug build: False
CUDA used to build PyTorch: 11.7
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.4 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.31

Python version: 3.8.10 (default, Nov 14 2022, 12:59:47)  [GCC 9.4.0] (64-bit runtime)
Python platform: Linux-5.15.0-50-generic-x86_64-with-glibc2.29
Is CUDA available: True
CUDA runtime version: 11.7.64
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: Tesla V100-DGXS-32GB
GPU 1: Tesla V100-DGXS-32GB
GPU 2: Tesla V100-DGXS-32GB
GPU 3: Tesla V100-DGXS-32GB

Nvidia driver version: 470.141.03
cuDNN version: Probably one of the following:
/usr/local/cuda-11.3/targets/x86_64-linux/lib/libcudnn.so.8.4.1
/usr/local/cuda-11.3/targets/x86_64-linux/lib/libcudnn_adv_infer.so.8.4.1
/usr/local/cuda-11.3/targets/x86_64-linux/lib/libcudnn_adv_train.so.8.4.1
/usr/local/cuda-11.3/targets/x86_64-linux/lib/libcudnn_cnn_infer.so.8.4.1
/usr/local/cuda-11.3/targets/x86_64-linux/lib/libcudnn_cnn_train.so.8.4.1
/usr/local/cuda-11.3/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8.4.1
/usr/local/cuda-11.3/targets/x86_64-linux/lib/libcudnn_ops_train.so.8.4.1
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

Versions of relevant libraries:
[pip3] mypy==0.991
[pip3] mypy-extensions==0.4.3
[pip3] numpy==1.24.1
[pip3] torch==1.13.1
[pip3] torchvision==0.15.0a0+d2d448c
[conda] Could not collect

I am not sure if this has to do with the installation procedure from my side or if this is a general thing. Any help would be appreciated!

Thanks
Lukas

[NicolasHug]

Hi @lukazso

[pip3] torch==1.13.1

I suspect this is the problem: you're actually not using the nightly version (which should look like 2.0.dev....

You used:

pip install --pre torch -f https://download.pytorch.org/whl/nightly/cu117/torch_nightly.html

But the official installation instruction suggest

pip3 install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cu117

Perhaps try these?

---

Side note: which prototype features are you interested in? If it's about the new transforms, please feel free to provide any feedback you may have in this thread #6753 ! It will help us provide the best user experience.

[lukazso]

Thanks @NicolasHug for the quick feedback! This was indeed the problem. I am also closely following the thread regarding the new transforms (really cool feature).

However, now I encounter the following:

>>> from torchvision.prototype import datapoints, transforms as T
/home/user/project/vision/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: libnvjpeg.so.11: cannot open shared object file: No such file or directory
  warn(f"Failed to load image Python extension: {e}")

This looks similar to me as #7104. I already installed the additional libraries:

sudo apt install libjpeg-turbo8 libjpeg-turbo8-dev \
libpng16-16 libpng-tools libpng-dev

Also, I reinstalled torchvision from the repo again after installing the additional libraries.

Is there still something wrong with my installation?

[NicolasHug]

nvjpeg is the nvidia lib for jpeg decoding (unrelated to libjpeg-turbo / libjpeg). I'm not sure why it's not present on your system, it could be another problem with your env. If you don't need GPU-decoding of jpegs (you probably don't), you should be able to bypass the problem by setting TORCHVISION_USE_NVJPEG=0 before building torchvision

[Subhankhan41]

Hi,

I tried using nightly as well but I am still struggling to use from torchvision.prototype import datapoints or simply from torchvision import datapoints. Any fix to it?