MONAI v1.3.0

[laraPPr]

• link to support ticket: #2023112260000863 + #2024020860000217
• website: https://monai.io/
• installation docs: https://docs.monai.io/en/latest/installation.html
• toolchain: foss/2022a
• easyblock to use: PythonPackage
• required dependencies:
  • https://github.com/Project-MONAI/MONAI/blob/dev/requirements-dev.txt
• optional dependencies:
  • ...
• notes:
  • ...
• effort: (TBD)
• other install methods
  • conda: yes (https://anaconda.org/conda-forge/monai)
  • container image: yes (https://docs.monai.io/en/latest/installation.html#from-dockerhub)
  • pre-built binaries (RHEL8 Linux x86_64): no
  • easyconfig outside EasyBuild: yes

[boegel]

see easybuilders/easybuild-easyconfigs#19598

[PetrKralCZ]

• {vis}[foss/2023a] torchvision v0.16.0 w/ CUDA 12.1.1 easybuilders/easybuild-easyconfigs#20493
• {vis}[foss/2023a] MONAI v1.3.0 w/ CUDA 12.1.1 easybuilders/easybuild-easyconfigs#20494

[PetrKralCZ]

Surprisingly MONAI (easybuilders/easybuild-easyconfigs#20494) build fails with:

/apps/gent/RHEL8/cascadelake-ib/software/pybind11/2.11.1-GCCcore-12.3.0/include/pybind11/detail/../cast.h:45:120: error: expected template-name before < token
         45 |     return caster.operator typename make_caster<T>::template cast_op_type<T>();
            |                                                                                                                        ^
      /apps/gent/RHEL8/cascadelake-ib/software/pybind11/2.11.1-GCCcore-12.3.0/include/pybind11/detail/../cast.h:45:120: error: expected identifier before < token
      /apps/gent/RHEL8/cascadelake-ib/software/pybind11/2.11.1-GCCcore-12.3.0/include/pybind11/detail/../cast.h:45:123: error: expected primary-expression before > token
         45 |     return caster.operator typename make_caster<T>::template cast_op_type<T>();
            |                                                                                                                           ^
      /apps/gent/RHEL8/cascadelake-ib/software/pybind11/2.11.1-GCCcore-12.3.0/include/pybind11/detail/../cast.h:45:126: error: expected primary-expression before ) token
         45 |     return caster.operator typename make_caster<T>::template cast_op_type<T>();
            |                                                                                                                              ^

The non-CUDA version works and these are the only differences:

--- /user/gent/453/vsc45304/easyconfigs/m/MONAI/MONAI-1.3.0-foss-2023a.eb       2024-02-13 14:26:27.200431393 +0100
+++ MONAI-1.3.0-foss-2023a-CUDA-12.1.1.eb       2024-05-07 10:21:18.970757232 +0200
@@ -2,6 +2,7 @@
 
 name = 'MONAI'
 version = '1.3.0'
+versionsuffix = '-CUDA-%(cudaver)s'
 
 homepage = 'https://monai.io/'
 description = """
@@ -19,16 +20,17 @@
 ]
 
 dependencies = [
+    ('CUDA', '12.1.1', '', SYSTEM),
     ('Python', '3.11.3'),
     ('Python-bundle-PyPI', '2023.06'),
     ('SciPy-bundle', '2023.07'),
-    ('PyTorch', '2.1.2'),
+    ('PyTorch', '2.1.2', versionsuffix),
     ('einops', '0.7.0'),
     ('ITK', '5.3.0'),
     ('NiBabel', '5.2.0'),
     ('scikit-image', '0.22.0'),
     ('tensorboard', '2.15.1'),
-    ('torchvision', '0.16.0'),
+    ('torchvision', '0.16.0', versionsuffix),
     ('tqdm', '4.66.1'),
     ('Pillow', '10.0.0'),
     ('openslide-python', '1.3.1'),

[boegel]

@lexming Does this happen to ring any bells for you?

[lexming]

The issue is not with MONAI but pybind11, I suggest to reinstall pybind11 in the cascadelakes. That installation might lack some patches, like easybuilders/easybuild-easyconfigs#19047

[boegel]

Thanks a lot for the help on this one @lexming!

@PetrKralCZ PRs merged, software installed, ready for cleanup & closing...