# Undefined symbol pytorch 1.7.0 and above **URL:** \ **Category:** Uncategorized\ **Created:** [July 13, 2021, 3:13pm UTC](https://discuss.pytorch.org/t/undefined-symbol-pytorch-1-7-0-and-above/126640 "2021-07-13T15:13:39Z")\ **Posts on this page:** 4\ **Page:** 1
**Author:** ![czotti](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/czotti/32/8031_2.png) [@czotti](https://discuss.pytorch.org/u/czotti)\ **Post date:** [July 13, 2021, 3:13pm UTC](https://discuss.pytorch.org/t/undefined-symbol-pytorch-1-7-0-and-above/126640/1 "2021-07-13T15:13:39Z")
I was working with PyTorch 1.5.0 for quite some time and decided to update to 1.9.0. System: Ubuntu 18.04 Kernel: 4.15.0-147-generic #151-Ubuntu SMP I use pyenv 2.0.3 to manage virtual environment and python versions. I use python 3.7.9 installed with pyenv. Make a new virtual environment with pyenv called XXXX. My pip env shows ```auto $ pip --version pip 20.1.1 from /home/clement/.pyenv/versions/3.7.9/envs/XXXX/lib/python3.7/site-packages/pip (python 3.7) $ pip uninstall torch torchvision torchaudio WARNING: Skipping torch as it is not installed. WARNING: Skipping torchvision as it is not installed. WARNING: Skipping torchaudio as it is not installed. $ python3 -c "import torch; torch. __version__" Traceback (most recent call last): File "", line 1, in ModuleNotFoundError: No module named 'torch' ``` Everything is clean with the new setup. Now install PyTorch 1.9.0 everything goes well no error message from pip ```auto pip3 install torch==1.9.0+cpu torchvision==0.10.0+cpu torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html ``` So everything looks fine but ```auto $ python3 -c "import torch; torch. __version__" Traceback (most recent call last): File "", line 1, in File "/home/clement/.pyenv/versions/3.7.9/envs/XXXX/lib/python3.7/site-packages/torch/ __init__.py", line 197, in from torch._C import * # noqa: F403 ImportError: /home/clement/.pyenv/versions/3.7.9/envs/XXXX/lib/python3.7/site-packages/torch/lib/libtorch_cpu.so: undefined symbol: _ZN3c1019UndefinedTensorImpl10_singletonE ``` I found that any version above 1.7.0 (included) give me this kind of error message. - 1.7.0+cpu: \_ZTIN3c1010TensorImplE - 1.7.1+cpu: \_ZTIN3c1010TensorImplE - 1.8.0+cpu: \_ZTIN3c1010TensorImplE - 1.9.0+cpu: \_ZN3c1019UndefinedTensorImpl10\_singletonE I also tried the “normal” version `pip install torch==1.7.0 torchvision==0.8.0 torchaudio==0.7.0` - 1.7.0: \_ZN3c107Warning4warnENS\_14SourceLocationERKSsb I don’t have any compatible GPU so maybe the above error is as expected? I also tried with the python 3.6.9 from the ubuntu 18.04 repository I have the same behavior as above. Furthermore, I tried to clean anything on my system that contains `find / -iname *torch*`. Honestly I don’t know where to look for this issue. Any version prior to 1.7.0 install and works without any issues. I use docker with Ubuntu 18.04, and I could install 1.7.1 and works with torch. I think this come from my workstation, but I have no clues. If someone knows where I can start to look at. ---
**Author:** ![ptrblck](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/ptrblck/32/1823_2.png) [@ptrblck](https://discuss.pytorch.org/u/ptrblck)\ **Post date:** [July 14, 2021, 8:53am UTC](https://discuss.pytorch.org/t/undefined-symbol-pytorch-1-7-0-and-above/126640/2 "2021-07-14T08:53:16Z")
Usually you would get this error message if you are trying to `import` custom extensions before importing PyTorch, which doesn’t seem to be the case here. I also assume that you don’t have any files, which could cause conflicts, in the current working directory. In that case, could you create a new virtual environment, install PyTorch there, and test it again? ---
**Author:** ![czotti](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/czotti/32/8031_2.png) [@czotti](https://discuss.pytorch.org/u/czotti)\ **Post date:** [July 14, 2021, 5:30pm UTC](https://discuss.pytorch.org/t/undefined-symbol-pytorch-1-7-0-and-above/126640/3 "2021-07-14T17:30:33Z")
No directory/file was named torch on my working folder, if it was the case I assume that torch==1.6.0 will also fail no ? Here are the steps I have done for the new venv: ```auto $ mkdir new_venv $ cd new_venv $ virtualenv -p python3 pytorch_test Running virtualenv with interpreter /home/clement/.pyenv/shims/python3 Using base prefix '/usr' New python executable in /home/clement/new_env/pytorch_test/bin/python3 Also creating executable in /home/clement/new_env/pytorch_test/bin/python Installing setuptools, pkg_resources, pip, wheel...done. $ source pytorch_test/bin/activate (pytorch_test) $ where python3 /home/clement/new_env/pytorch_test/bin/python3 /usr/bin/python3 (pytorch_test) $ pip freeze pkg-resources==0.0.0 (pytorch_test) $ pip install torch==1.7.0+cpu torchvision==0.8.0 torchaudio==0.7.0 -f https://download.pytorch.org/whl/torch_stable.html (pytorch_test) $ python3 -c "import torch; print(torch. __version__ )" Traceback (most recent call last): File "", line 1, in File "/home/clement/new_env/pytorch_test/lib/python3.6/site-packages/torch/ __init__.py", line 190, in from torch._C import * ImportError: /home/clement/new_env/pytorch_test/lib/python3.6/site-packages/torch/lib/libtorch_cpu.so: undefined symbol: _ZTIN3c1010TensorImplE (pytorch_test) $ pip uninstall torch torchvision torchaudio (pytorch_test) $ (pytorch_test) $ pip install torch==1.6.0+cpu torchvision==0.7.0+cpu -f https://download.pytorch.org/whl/torch_stable.html (pytorch_test) $ python3 -c "import torch; print(torch. __version__ )" 1.6.0+cpu ``` ---
**Author:** ![czotti](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/czotti/32/8031_2.png) [@czotti](https://discuss.pytorch.org/u/czotti)\ **Post date:** [August 6, 2021, 6:54pm UTC](https://discuss.pytorch.org/t/undefined-symbol-pytorch-1-7-0-and-above/126640/4 "2021-08-06T18:54:30Z")
The only way I found to make it work is to use a docker container with the required PyTorch version in it.