skip CuPy 14.1.0

[jameslamb]

Contributes to rapidsai/build-planning#284

To reduce the risk of users encountering a known runtime issue with cupy and torch, RAPIDS 26.06 is taking on a requirement cupy!=14.1.0.

See the linked issue for details.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[jameslamb]

All CUDA 13 wheel-testing jobs are failing like this:

=========================== short test summary info ============================
FAILED python/kvikio/tests/test_benchmarks.py::test_single_node_io[cufile] - assert 1 == 0
FAILED python/kvikio/tests/test_benchmarks.py::test_single_node_io[posix] - assert 1 == 0
FAILED python/kvikio/tests/test_benchmarks.py::test_single_node_io[cufile-mfma] - assert 1 == 0
FAILED python/kvikio/tests/test_benchmarks.py::test_single_node_io[cufile-mf] - assert 1 == 0
FAILED python/kvikio/tests/test_benchmarks.py::test_single_node_io[cufile-ma] - assert 1 == 0
============ 5 failed, 1512 passed, 13 skipped in 105.94s (0:01:45) ============

(build link)

CUDA 12 jobs are passing.

I'll go ask for some help on that.

[TomAugspurger]

For reference, we apparently had a typo in our cuda.core.system usage. It's system.CUDA_BINDINGS_NVML_IS_COMPATIBLE, not system.CUDA_BINDINGS_IS_NVML_COMPATIBLE. It only showed up in the 13.x builds because that's the only one that pulled cuda-core into the environment. I haven't dug any further than confirming that it's always been system.CUDA_BINDINGS_NVML_IS_COMPATIBLE.

[jakirkham]

Thanks James for the CuPy dependency fix! Also thanks Tom for fixing the test issues that popped up 🙏

LGTM