`cuda.bindings` 12.9.0 & 11.8.7

[leofang]

Tasks for cuda-core and cuda-bindings patch release

• File an internal nvbug to communicate test plan & release schedule with QA
• Ensure all pending PRs are reviewed, tested, and merged
• Check (or update if needed) the dependency requirements
• Bump the version
• Create a public rc tag
• Point QA to fetch public artifacts (wheels) from the GHA run ID, example: gh run download 12323257563 -p "cuda-core*" -R NVIDIA/cuda-python
• Wait for QA reports and fix any issues found
• Finalize the doc update, including release notes ("Note: Touching docstrings/type annotations in code is OK during code freeze, apply your best judgement!")
• Update the docs for the new version
• Create a public release tag
• If any code change happens, rebuild the wheels from the new tag
• Update the conda recipe & release conda packages
• Upload conda packages to nvidia channel
• Upload wheels to PyPI
• Post-release QA
• Finalize the announcement update
• Send out the announcement internally
• Send out the announcement externally (GitHub Release -> Announcement)

[leofang]

Before releasing: We should manually test this use case #274 (comment). @rwgk let's discuss on Monday. It should be very quick.

[leofang]

Before releasing: We should manually test this use case #274 (comment). @rwgk let's discuss on Monday. It should be very quick.

We tested manually with rmm-cu12 25.04 (built against 12.8.0) and cuda-bindings 12.9.0, no issues found (i.e. import rmm works).

[rwgk]

For completeness, this is what we did exactly:

1. Ensure that there is no problem using cuda-bindings 12.9.0:

conda create -y -n test_rmm128_with_bindings129 python=3.12
conda activate test_rmm128_with_bindings129
pip install numpy
conda install -y gh
gh run download 14822788475 -p cuda-bindings-python312-cuda12.9.0-linux-64-* -R NVIDIA/cuda-python
pip install cuda-bindings-python312-cuda12.9.0-linux-64-dcb343cce96b0a55ddd446af69d46bbc56b26a21/cuda_bindings-12.9.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
pip install --no-deps rmm-cu12 rapids-logger librmm-cu12

(test_rmm128_with_bindings129) 0db6015-lcedt.nvidia.com:~ $ python
Python 3.12.10 | packaged by conda-forge | (main, Apr 10 2025, 22:21:13) [GCC 13.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import rmm
>>> rmm.DeviceBuffer(size=100)
<rmm.pylibrmm.device_buffer.DeviceBuffer object at 0x7062807d5980>
>>>

2. Reproduce the old error in the same environment (to ensure that our approach is conclusive):

pip uninstall --yes cuda-bindings
pip install "cuda-python<12.6.2"

(test_rmm128_with_bindings129) 0db6015-lcedt.nvidia.com:~ $ python
Python 3.12.10 | packaged by conda-forge | (main, Apr 10 2025, 22:21:13) [GCC 13.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import rmm
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/rgrossekunst/miniforge3/envs/test_rmm128_with_bindings129/lib/python3.12/site-packages/rmm/__init__.py", line 26, in <module>
    from rmm import mr
  File "/home/rgrossekunst/miniforge3/envs/test_rmm128_with_bindings129/lib/python3.12/site-packages/rmm/mr.py", line 14, in <module>
    from rmm.pylibrmm.memory_resource import (
  File "/home/rgrossekunst/miniforge3/envs/test_rmm128_with_bindings129/lib/python3.12/site-packages/rmm/pylibrmm/__init__.py", line 15, in <module>
    from .device_buffer import DeviceBuffer
  File "device_buffer.pyx", line 1, in init rmm.pylibrmm.device_buffer
  File "memory_resource.pyx", line 32, in init rmm.pylibrmm.memory_resource
ModuleNotFoundError: No module named 'cuda.bindings'
>>>

Just cleanup:

conda deactivate
conda remove -y -n test_rmm128_with_bindings129 --all
rm -rf cuda-bindings-python312-cuda12.9.0-linux-64-dcb343cce96b0a55ddd446af69d46bbc56b26a21