error when install apex and compile normalSpeed

[minwang-ai]

Hi all,

I create a conda .yml file and another requirements.txt for creating conda environment because I followed the tutorial of my administrator of GPU cluster to use conda install at first and only use pip when the packages are not available on conda.

However, I got some errors when I install apex as the following:

After that, I add apex into .yml file and install NormalSpeed but I got the same error as this issue.

Could you help me fix it?

These are my .yml and .txt files .

Best, Min

[ethnhe]

• For the apex installation problem please refer to the apex project, try to raise an issue or search for solutions in the past issues there.
• For the NormalSpeed issue try to install opencv3 from the source code as mention in the issue.

[minwang-ai]

• For the apex installation problem please refer to the apex project, try to raise an issue or search for solutions in the past issues there.

Hi Yisheng, Thank you for your reply. I have already raised an issue and searched for solutions in the past issues for apex. Can I simply install apex by conda?

The problem of NormalSpeed disappeared automatically when I install apex by conda.

[ethnhe]

Sorry that I didn't try to install apex by conda. You can have a try, see if it works well, and update the issue.

[minwang-ai]

Sorry that I didn't try to install apex by conda. You can have a try, see if it works well, and update the issue.

No it shows the same error - lack of cpp extension. I tried to install the latest torch version and installed apex successfully.

conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch

git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

and then install other packages...