Here you go :)

error: Cannot build '/nix/store/dgxvh459r8lqlcpli60vjdahqm03ybn6-python3.13-flash-attention-2.8.3.drv'.
       Reason: builder failed with exit code 1.
       Output paths:
         /nix/store/13nyvan7jp2abknfr56plx3fyn89zp7f-python3.13-flash-attention-2.8.3
         /nix/store/f0z9q35rz8xh68q6qcnfiz6mjla8jdzl-python3.13-flash-attention-2.8.3-dist
       Last 25 log lines:
       > Search for `cudaErrorIllegalAddress' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
       > CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
       > For debugging consider passing CUDA_LAUNCH_BLOCKING=1
       > Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
       > FAILED tests/test_rotary.py::test_rotary_emb_varlen_func[True-True-0.5-int-dtype1] - torch.AcceleratorError: CUDA error: an illegal memory access was encountered
       > Search for `cudaErrorIllegalAddress' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
       > CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
       > For debugging consider passing CUDA_LAUNCH_BLOCKING=1
       > Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
       > FAILED tests/test_rotary.py::test_rotary_emb_varlen_func[True-True-0.5-Tensor-dtype0] - torch.AcceleratorError: CUDA error: an illegal memory access was encountered
       > Search for `cudaErrorIllegalAddress' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
       > CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
       > For debugging consider passing CUDA_LAUNCH_BLOCKING=1
       > Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
       > FAILED tests/test_rotary.py::test_rotary_emb_varlen_func[True-True-0.5-Tensor-dtype1] - torch.AcceleratorError: CUDA error: an illegal memory access was encountered
       > Search for `cudaErrorIllegalAddress' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
       > CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
       > For debugging consider passing CUDA_LAUNCH_BLOCKING=1
       > Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
       > FAILED tests/test_rotary.py::test_compilation_count - torch.AcceleratorError: CUDA error: an illegal memory access was encountered
       > Search for `cudaErrorIllegalAddress' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.
       > CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
       > For debugging consider passing CUDA_LAUNCH_BLOCKING=1
       > Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
       > = 197981 failed, 155167 passed, 213305 skipped, 19 warnings in 72020.06s (20:00:20) =
       For full logs, run:
         nix log /nix/store/dgxvh459r8lqlcpli60vjdahqm03ybn6-python3.13-flash-attention-2.8.3.drv

The full log is 1.9GB, I can also share it with you if its useful

In reply to @glepage:matrix.org
I realized both onnxruntime and opencv fail to build with cudaPackages_13_2.

No, it's a compilation error:

/nix/store/9bvirxizyiq3pg5rmgpz2f6aw6y7c1fm-cuda13.2-cuda_cccl-13.2.27/include/cub/device/device_transform.cuh:44:111: error: global qualification of class name is invalid before ':' token
   44 | struct ::cuda::proclaims_copyable_arguments<CUB_NS_QUALIFIER::detail::__return_constant<T>> : ::cuda::std::true_type
      |                                                                                                               ^
/nix/store/9bvirxizyiq3pg5rmgpz2f6aw6y7c1fm-cuda13.2-cuda_cccl-13.2.27/include/cub/device/device_transform.cuh:44:111: error: expected '{' before ':' token
make[2]: *** [CMakeFiles/onnxruntime_providers_cuda.dir/build.make:8382: CMakeFiles/onnxruntime_providers_cuda.dir/build/source/onnxruntime/contrib_ops/cuda/math/bias_softmax_impl.cu.o] Error 1
make[2]: *** Waiting for unfinished jobs....
In file included from /nix/store/54yzskzwmapls2awvih149wyzci717gw-onnx-src/onnx/version_converter/convert.cc:5:
/nix/store/54yzskzwmapls2awvih149wyzci717gw-onnx-src/onnx/version_converter/convert.h: In constructor 'onnx::version_conversion::DefaultVersionConverter::DefaultVersionConverter()':
/nix/store/54yzskzwmapls2awvih149wyzci717gw-onnx-src/onnx/version_converter/convert.h:110:3: note: variable tracking size limit exceeded with '-fvar-tracking-assignments', retrying without
  110 |   DefaultVersionConverter() {
      |   ^~~~~~~~~~~~~~~~~~~~~~~
/nix/store/9bvirxizyiq3pg5rmgpz2f6aw6y7c1fm-cuda13.2-cuda_cccl-13.2.27/include/cub/device/device_transform.cuh:44:111: error: global qualification of class name is invalid before ':' token
   44 | struct ::cuda::proclaims_copyable_arguments<CUB_NS_QUALIFIER::detail::__return_constant<T>> : ::cuda::std::true_type
      |                                                                                                               ^
/nix/store/9bvirxizyiq3pg5rmgpz2f6aw6y7c1fm-cuda13.2-cuda_cccl-13.2.27/include/cub/device/device_transform.cuh:44:111: error: expected '{' before ':' token
make[2]: *** [CMakeFiles/onnxruntime_providers_cuda.dir/build.make:8562: CMakeFiles/onnxruntime_providers_cuda.dir/build/source/onnxruntime/contrib_ops/cuda/moe/ft_moe/moe_kernel.cu.o] Error 1

Here's a good one 🥲🐸

https://github.com/NixOS/nixpkgs/pull/525504
cc Kevin Mittman (UTC-7)

Here's a good one 🥲🐸

https://github.com/NixOS/nixpkgs/pull/525504
cc Kevin Mittman (UTC-7)

(Maybe I'm just holding it wrong and I badly packaged apex)