Uploaded image for project: 'AI Platform Core Components'
  1. AI Platform Core Components
  2. AIPCC-7369

terratorch==1.1.1 requires old albucore, which conflicts in numpy<2.0, jsonargparse

    • Icon: Bug Bug
    • Resolution: Unresolved
    • Icon: Undefined Undefined
    • None
    • rhoai-3.2
    • Development Platform
    • None
    • Important

      pDescription of problem:

      The latest releases of rasterio, pyproj, shapely, torchgeo, terratorch, timm, mmsegmentation are building fine with Torch 2.9.0. However Fromager is unable to create an installable set of packages. terratorch==1.1.1 depends on an old version of albucore, which indirectly requires numpy<2.0. Other packages requires numpy>2.0

      Version numbers (base image, wheels, builder, etc):

      Builder 24.8.0
      terratorch==1.1.1
      albucore==0.0.16 and 0.0.34
      albumentations==1.4.10

      Steps to Reproduce:

      https://gitlab.com/redhat/rhel-ai/core/testcollections/selfservice/-/jobs/12171173440

      Actual results:

      10:58:52 ERROR jsonargparse: no single version meets all requirements
      10:58:52 ERROR numpy: no single version meets all requirements
      ...
      albucore
        0.0.16
          albucore==0.0.16 matches ['0.0.16']
            terratorch==1.1.1
        0.0.34
          albucore>=0.0.11 matches ['0.0.16', '0.0.34']
            albumentations==1.4.10
        * albucore==0.0.16 usable by all consumers
      jsonargparse
        4.35.0
          jsonargparse<=4.35.0 matches ['4.35.0']
            terratorch==1.1.1
        4.43.0
          jsonargparse[signatures]>=4.25 matches ['4.43.0', '4.35.0']
            torchgeo==0.7.2
          jsonargparse[typing-extensions]; extra == "signatures" matches ['4.43.0', '4.35.0']
            jsonargparse==4.43.0
          jsonargparse matches ['4.43.0', '4.35.0']
            terratorch==1.1.1
          jsonargparse[jsonnet,signatures]<5.0,>=4.39.0; extra == "pytorch-extra" matches ['4.43.0']
            lightning==2.5.6
        * No single version of jsonargparse meets all requirements
      numpy
        1.26.4
          numpy<2,>=1.24.4 matches ['1.26.4']
            albumentations==1.4.10
        2.3.5
          numpy>=1.24 matches ['2.3.5', '1.26.4']
            scikit-image==0.25.2
            rasterio==1.4.3
            geopandas==1.1.1
            albucore==0.0.16
          numpy>=1.21 matches ['2.3.5', '1.26.4']
            shapely==2.1.2
          numpy matches ['2.3.5', '1.26.4']
            imageio==2.37.2
            pycocotools==2.0.10
            lightly-utils==0.0.2
            mmsegmentation==1.2.2
            pythran==0.18.1
            tifffile==2025.10.16
            tensorboardx==2.6.4
            pyogrio==0.11.1
            torchvision==0.24.0
          numpy>=1.18.1 matches ['2.3.5', '1.26.4']
            lightly==1.5.22
          numpy>1.20.0 matches ['2.3.5', '1.26.4']
            torchmetrics==1.8.2
          numpy>=1.25 matches ['2.3.5', '1.26.4']
            contourpy==1.3.3
          numpy>=1.23 matches ['2.3.5', '1.26.4']
            matplotlib==3.10.7
          numpy>=1.23.2 matches ['2.3.5', '1.26.4']
            torchgeo==0.7.1
            torchgeo==0.7.2
          numpy>=1.26.0; python_version >= "3.12" matches ['2.3.5', '1.26.4']
            pandas==2.3.3
            opencv-python-headless==4.11.0.86
          numpy>=1.19.3 matches ['2.3.5', '1.26.4']
            segmentation-models-pytorch==0.5.0
          numpy>=1.21.2; python_version >= "3.10" matches ['2.3.5', '1.26.4']
            opencv-python-headless==4.11.0.86
          numpy>=1.17.0; python_version >= "3.7" matches ['2.3.5', '1.26.4']
            opencv-python-headless==4.11.0.86
          numpy>=1.17.3; python_version >= "3.8" matches ['2.3.5', '1.26.4']
            opencv-python-headless==4.11.0.86
          numpy>=1.23.5; python_version >= "3.11" matches ['2.3.5', '1.26.4']
            opencv-python-headless==4.11.0.86
          numpy>=1.19.3; python_version >= "3.9" matches ['2.3.5', '1.26.4']
            opencv-python-headless==4.11.0.86
          numpy>=1.24.4 matches ['2.3.5', '1.26.4']
            albucore==0.0.34
          numpy<2.6,>=1.25.2 matches ['2.3.5', '1.26.4']
            scipy==1.16.3
          numpy>=1.22.0 matches ['2.3.5', '1.26.4']
            scikit-learn==1.7.2
          numpy>=1.21.2 matches ['2.3.5', '1.26.4']
            h5py==3.15.1
          numpy>=2 matches ['2.3.5']
            rioxarray==0.20.0
          numpy>=1.26 matches ['2.3.5', '1.26.4']
            xarray==2025.11.0
          numpy>=1.12.0 matches ['2.3.5', '1.26.4']
            tensorboard==2.20.0
          numpy>=1.17 matches ['2.3.5', '1.26.4']
            bitsandbytes==0.48.2
        * No single version of numpy meets all requirements
      torchgeo
        0.7.1
          torchgeo<0.7.2,>=0.7.0 matches ['0.7.1']
            terratorch==1.1.1
        0.7.2
        * torchgeo==0.7.1 usable by all consumers    

      Expected results:

          

      Additional info:

              Unassigned Unassigned
              cheimes@redhat.com Christian Heimes
              Antonio's Team
              Votes:
              0 Vote for this issue
              Watchers:
              3 Start watching this issue

                Created:
                Updated: