I cannot use the pytorch that was built successfully from source: (DLL) initialization routine failed. Error loading caffe2_detectron_ops_gpu.dll

@peterjc123
I do not get along with your suggested link. Why do I have to install mkl like that, if I can simply install it with conda?

It seems to be better to follow GitHub - pytorch/pytorch: Tensors and Dynamic neural networks in Python with strong GPU acceleration and use the dependency installer there (and where mkl is included anyway?):

conda install numpy ninja pyyaml mkl mkl-include setuptools cmake cffi typing_extensions future six requests

####

Aside from that, there is a mistake in your stated link.

REM Download MKL files

curl https://s3.amazonaws.com/ossci-windows/mkl_2020.0.166.7z -k -O
7z x -aoa mkl_2018.2.185.7z -omkl

It works with
7z x -aoa mkl_2020.0.166.7z -omkl
since that is what the curl command is downloading as a 7z.

Anyway, I have now succeeded in using the two commands, having now C:\Users\Admin\mkl with the necessary files in it, but I do not need the directory at all, since I can simply install mkl with conda, see above.

####

I do not know now how to install from source with MKL. Looking at your link, under Windows FAQ — PyTorch 2.1 documentation, I find:

conda install -c peterjc123 vc vs2017_runtime
conda install mkl_fft intel_openmp numpy mkl

But that does not seem to help me, first, I have vs2019, then I want to install pytorch with mkl, and I find a wide range of commands that do not seem to help.

Because pip install intel-openmp does not work, I use from your link:

conda install -c defaults intel-openmp -f

Which seems to be needed for installing pytorch with mkl, as you say: “Please build with MKL so Intel OpenMP will be used.”

Please give me a hint what I shall do next to get pytorch built with MKL, thank you.