Add py-torch (in some non-cuda version)

Egidio Falotico mentioned NRP has problems integrating py-torch "into the platform". Let's see what happens…

mentioned in merge request !34 (closed)

added 1 commit

• f3b3971f - Add py-torch (in some non-cuda version)

Compare with previous version

added 7 commits

• f3b3971f...31c52815 - 6 commits from branch technical-coordination/project-internal/devops/platform:master
• 291dcb66 - Add py-torch (in some non-cuda version)

Compare with previous version

mentioned in commit 68038cbe

merged

The build of py-torch failed.
Please check it.
https://gitlab.ebrains.eu/technical-coordination/project-internal/devops/platform/ebrains-spack-builds/-/jobs/20053

Hmm, looks like we need an older protobuf for PyTorch (the function signature that fails has been changed after 3.5… cf.

https://github.com/protocolbuffers/protobuf/blob/3.5.x/src/google/protobuf/io/coded_stream.cc#L176

vs.

https://github.com/protocolbuffers/protobuf/blob/3.6.x/src/google/protobuf/io/coded_stream.cc#L176

Probably the py-torch package needs a constraint on the protobuf version… looking into it.

Ok, let's try a simpler workaround → disabling onnx support as this introduces the failing code…