demo.py segment fault

[mycprotein]

I have settled the environment like this:
sudo apt install openmpi-common openmpi-bin
pip install torch == 1.12
pip install oneccl_bind_pt==1.12.0 -f https://developer.intel.com/ipex-whl-stable
source <oneccl_bindings_for_pytorch_path>/env/setvars.sh
Then I execute the demo.py file using python demo.py or using mpirun, but in both way I get a segmentation fault.
I think maybe there are some misuses, but I cannot find detailed documents. I wonder how can I make it work?

[mycprotein]

I have checked that I use torch==1.12.0.

[zhouyu5]

I had the exact same problem, hope it will be fixed soon, thanks.

[zxd1997066]

Since for release 1.12.0, pytorch has two version: with +cpu or not. You may download torch-1.12.0+cpu from https://download.pytorch.org/whl/cpu/torch/ and it can work.

[zhouyu5]

Thanks for replying @zxd1997066

Actually we are using pytorch with cpu version from OneAPI docker images, say:

docker pull intel/oneapi-aikit:2022.3.0-devel-ubuntu18.04

So I can gurantee you that we are using a CPU version Pytorch.

In the pictures above, you show that it works fine in single node, but I suggest you try to run it distributedly, that is, to run it by using the MPI tool set, just as illustrated in the Github homepage, using the command like:

mpirun -n <N> -ppn <PPN> -f <hostfile> python example.py

Me and my colleagues got the same segmentation fault error. Could you please have a try?

Thank you very much!

[zxd1997066]

I am sorry, I can not reproduce it.

[zigzagcai]

Hello, I have tested with docker pull intel/oneapi-aikit:2022.3.1-devel-ubuntu18.04. The same error accured.

[zigzagcai]

I am sorry, I can not reproduce it.

Hi bro, which docker base image are you using? I have tried with oneapi-aikit base image but it reported error.

[zxd1997066]

I am sorry, I can not reproduce it.

Hi bro, which docker base image are you using? I have tried with oneapi-aikit base image but it reported error.

I just run it in conda, so I guess maybe there is something wrong with that docker, and this needs further investigate.