ONNXRuntime won't build

[kpedro88]

Description

Running build.py with onnxruntime_backend (links provide exact commit hashes), the ONNXRuntime build fails with:

#13 3412.7 /usr/local/cuda/targets/x86_64-linux/include/cccl/cub/device/device_transform.cuh:45:97: error: global qualification of class name is invalid before ‘:’ token
#13 3412.7    45 | struct ::cuda::proclaims_copyable_arguments<CUB_NS_QUALIFIER::detail::__return_constant<T>> : ::cuda::std::true_type
#13 3412.7       |                                                                                                 ^
#13 3412.7 /usr/local/cuda/targets/x86_64-linux/include/cccl/cub/device/device_transform.cuh:45:97: error: expected ‘{’ before ‘:’ token
#13 3412.8 gmake[2]: *** [CMakeFiles/onnxruntime_providers_cuda.dir/build.make:7222: CMakeFiles/onnxruntime_providers_cuda.dir/workspace/onnxruntime/onnxruntime/contrib_ops/cuda/math/bias_softmax_impl.cu.o] Error 255

This looks related to microsoft/onnxruntime#28023.

Triton Information
What version of Triton are you using? See above

Are you using the Triton container or did you build it yourself? Built myself

To Reproduce

See above. I am just going to pin to an older version of the onnxruntime_backend repo that works (r26.04 with --ort-version 1.24.4, for the record), but I'm reporting this here in case someone has time to dissect it further.

Expected behavior

The build should complete successfully.