Fix vision install for all files

[tinglvv]

2026-04-02T12:26:57.0297276Z C:/actions-runner/_work/_temp/conda_environment_23898900289/lib/site-packages/torch/include\c10/cuda/CUDACachingAllocator.h(105): error: invalid combination of type specifiers
2026-04-02T12:26:57.1656746Z     StreamSegmentSize(cudaStream_t s, bool char , size_t sz)
2026-04-02T12:26:57.2674865Z                                            ^
2026-04-02T12:26:57.4435273Z 
2026-04-02T12:26:57.5852644Z C:/actions-runner/_work/_temp/conda_environment_23898900289/lib/site-packages/torch/include\c10/cuda/CUDACachingAllocator.h(106): error: type name is not allowed
2026-04-02T12:26:57.7023246Z         : stream(s), is_small_pool(char ), total_size(sz) {}
2026-04-02T12:26:57.8187869Z                                    ^
2026-04-02T12:26:57.9261140Z 
2026-04-02T12:26:58.7092980Z 2 errors detected in the compilation of "C:/actions-runner/_work/vision/vision/pytorch/vision/torchvision/csrc/ops/cuda/nms_kernel.cu".

Previous fix #9464 only applies to the .cpp files compiled directly by cl.exe, not .cu files via nvcc. Need to change it in the setup.py.

cc @atalman

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/9465

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures

As of commit f88e169 with merge base 4e58149 ():

NEW FAILURES - The following jobs have failed:

• Tests / unittests-linux (3.14, linux.12xlarge, cpu) / linux-job (gh)
  RuntimeError: Command docker exec -t eecf665d91ffee022be6126b84d674a92e8f26930e3ce619d8df3f6094ad74d3 /exec failed with exit code 1
• Tests / unittests-windows (3.14, windows.4xlarge, cpu) / windows-job (gh)
  ModuleNotFoundError: No module named 'packaging.version'

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[atalman]

lgtm

[jovan2009]

2026-04-02T12:26:57.0297276Z C:/actions-runner/_work/_temp/conda_environment_23898900289/lib/site-packages/torch/include\c10/cuda/CUDACachingAllocator.h(105): error: invalid combination of type specifiers
2026-04-02T12:26:57.1656746Z     StreamSegmentSize(cudaStream_t s, bool char , size_t sz)
2026-04-02T12:26:57.2674865Z                                            ^
2026-04-02T12:26:57.4435273Z 
2026-04-02T12:26:57.5852644Z C:/actions-runner/_work/_temp/conda_environment_23898900289/lib/site-packages/torch/include\c10/cuda/CUDACachingAllocator.h(106): error: type name is not allowed
2026-04-02T12:26:57.7023246Z         : stream(s), is_small_pool(char ), total_size(sz) {}
2026-04-02T12:26:57.8187869Z                                    ^
2026-04-02T12:26:57.9261140Z 
2026-04-02T12:26:58.7092980Z 2 errors detected in the compilation of "C:/actions-runner/_work/vision/vision/pytorch/vision/torchvision/csrc/ops/cuda/nms_kernel.cu".

Previous fix #9464 only applies to the .cpp files compiled directly by cl.exe, not .cu files via nvcc. Need to change it in the setup.py.

cc @atalman

I encounter the same error, I reported it here: pytorch/pytorch#179005 . The issue seems to be with PyTorch itself. This PR pytorch/pytorch#179100 solves it for me, not only for TorchVision but also for when I try to build other projects.

[atalman]

HI @jovan2009 the issue is different now:

C:\actions-runner\_work\vision\vision>python pytorch/vision/test/smoke_test.py   || exit /b 1 
Traceback (most recent call last):
  File "C:\actions-runner\_work\vision\vision\pytorch\vision\test\smoke_test.py", line 8, in <module>
    import torch
  File "C:\actions-runner\_work\_temp\conda_environment_23955132613\lib\site-packages\torch\__init__.py", line 3029, in <module>
    import torch._native
  File "C:\actions-runner\_work\_temp\conda_environment_23955132613\lib\site-packages\torch\_native\__init__.py", line 8, in <module>
    from . import cutedsl_utils, ops, registry, triton_utils
  File "C:\actions-runner\_work\_temp\conda_environment_23955132613\lib\site-packages\torch\_native\cutedsl_utils.py", line 6, in <module>
    from packaging.version import Version
ModuleNotFoundError: No module named 'packaging'

[jovan2009]

HI @jovan2009 the issue is different now:

C:\actions-runner\_work\vision\vision>python pytorch/vision/test/smoke_test.py   || exit /b 1 
Traceback (most recent call last):
  File "C:\actions-runner\_work\vision\vision\pytorch\vision\test\smoke_test.py", line 8, in <module>
    import torch
  File "C:\actions-runner\_work\_temp\conda_environment_23955132613\lib\site-packages\torch\__init__.py", line 3029, in <module>
    import torch._native
  File "C:\actions-runner\_work\_temp\conda_environment_23955132613\lib\site-packages\torch\_native\__init__.py", line 8, in <module>
    from . import cutedsl_utils, ops, registry, triton_utils
  File "C:\actions-runner\_work\_temp\conda_environment_23955132613\lib\site-packages\torch\_native\cutedsl_utils.py", line 6, in <module>
    from packaging.version import Version
ModuleNotFoundError: No module named 'packaging'

I think that is a completely different issue. Cuda 13.2 builds are still showing the same CUDACachingAllocator.h error:
https://github.com/pytorch/vision/actions/runs/24030647778/job/70078514526

https://github.com/pytorch/vision/actions/runs/23955132613/job/70051911372

My impression is that Cuda 13.2 builds are failing earlier, during building. All the others are failing a test after building successfully.

@atalman

[tinglvv]

Yes, the actual error that should be blocking is the CUDACachingAllocator.h(106): error: type name is not allowed. The other errors ModuleNotFoundError: No module named 'packaging' we have seen this before, I think it is infra issue.

[tinglvv]

Build still failing in https://github.com/pytorch/vision/actions/runs/23955132613/job/70243562557 since https://github.com/pytorch/pytorch/pull/179100/changes is not in nightly branch yet.

[github-actions]

Hey @atalman!

You merged this PR, but no labels were added.
The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py