Add Windows 13.0 domain builds

[tinglvv]

pytorch/pytorch#159779

cc @atalman

[vercel]

@tinglvv is attempting to deploy a commit to the Meta Open Source Team on Vercel.

A member of the Team first needs to authorize it.

[atalman]

@tinglvv I believe need to run python -m tools.tests.test_generate_binary_build_matrix --update-reference-files

[tinglvv]

Windows nightly build is still not available as of 9/3 as the commit does not include the windows build PR yet pytorch/pytorch@aa0545f, therefore causing the build failure. But we should be good to merge this PR despite the failure.

[atalman]

This is the error:

>>> from torch.utils.cpp_extension import BuildExtension
Windows fatal exception: access violation

Current thread 0x000015d8 (most recent call first):
  File "C:\Jenkins\Miniconda3\envs\py310\lib\site-packages\torch\cuda\__init__.py", line 182 in is_available
  File "C:\Jenkins\Miniconda3\envs\py310\lib\site-packages\torch\utils\cpp_extension.py", line 116 in _find_cuda_home
  File "C:\Jenkins\Miniconda3\envs\py310\lib\site-packages\torch\utils\cpp_extension.py", line 235 in <module>
  File "<frozen importlib._bootstrap>", line 241 in _call_with_frames_removed
  File "<frozen importlib._bootstrap_external>", line 883 in exec_module
  File "<frozen importlib._bootstrap>", line 688 in _load_unlocked
  File "<frozen importlib._bootstrap>", line 1006 in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 1027 in _find_and_load
  File "<stdin>", line 1 in <module>

Digging deeper:

>>> import torch
>>> import faulthandler
>>> faulthandler.enable()
>>> torch.cuda.is_available()
Windows fatal exception: access violation

Current thread 0x00001980 (most recent call first):
  File "C:\Jenkins\Miniconda3\envs\py310\lib\site-packages\torch\cuda\__init__.py", line 182 in is_available
  File "<stdin>", line 1 in <module>

I believe this should work no matter cpu or gpu machine:

>>> import torch
/home/ec2-user/github/variant-repack/.venv/lib/python3.13/site-packages/torch/_subclasses/functional_tensor.py:279: UserWarning: Failed to initialize NumPy: No module named 'numpy' (Triggered internally at /pytorch/torch/csrc/utils/tensor_numpy.cpp:81.)
  cpu = _conversion_method_template(device=torch.device("cpu"))
>>> torch.__version__
'2.8.0+cu128'
>>> torch.cuda.is_available()
False

Works with cu128 build on Windows:

Python 3.10.18 | packaged by Anaconda, Inc. | (main, Jun  5 2025, 13:08:55) [MSC v.1929 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
False

Link to nightly smoke tests: https://github.com/pytorch/pytorch/actions/runs/17486852873/job/49681168445

[atalman]

Digging deeper looks like failure when:

(py310) C:\actions-runner\_work\vision>python
Python 3.10.18 | packaged by Anaconda, Inc. | (main, Jun  5 2025, 13:08:55) [MSC v.1929 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> hasattr(torch._C, "_cuda_getDeviceCount")
True
>>> torch._C._cuda_getDeviceCount()

[atalman]

Error from windows log

Faulting application name: python.exe, version: 3.10.18150.1013, time stamp: 0x6841977c
Faulting module name: c10_cuda.dll, version: 0.0.0.0, time stamp: 0x68baa4b4
Exception code: 0xc0000005
Fault offset: 0x00000000000015d3
Faulting process id: 0x1908
Faulting application start time: 0x01dc1e9d4857191a
Faulting application path: C:\Jenkins\Miniconda3\envs\py310\python.exe
Faulting module path: C:\Jenkins\Miniconda3\envs\py310\lib\site-packages\torch\lib\c10_cuda.dll
Report Id: 98017187-f664-4643-9a6e-5bbd05d1084c
Faulting package full name: 
Faulting package-relative application ID: 

[atalman]

Same issue with GPU machine and Nvidia driver:

 NVIDIA-SMI 528.89       Driver Version: 528.89       CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name            TCC/WDDM | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Tesla T4           TCC   | 00000000:00:1E.0 Off |                    0 |
| N/A   35C    P8    11W /  70W |      0MiB / 15360MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+