[4031421040] Upstream Reporter: Lalatendu Mohanty
Upstream issue status: Open
Upstream description:

Notebooks uses `tensorflow[and-cuda]` on PyPI to pull in `nvidia-*` pip packages for CUDA support. On AIPCC Python Index, this extra doesn't exist ? the wheel already expects system CUDA. This forces pyproject.toml to differ between midstream and downstream.

If the base-containers CUDA image provides the necessary CUDA/cuDNN libs at the system level, then plain `tensorflow~=X.Y.0` (no extra) should work on both PyPI and AIPCC.

*What to do:*

• Verify the CUDA base images provide the libs that tensorflow needs (so the `[and-cuda]` extra pulling `nvidia-*` pip packages is unnecessary)
• Document this as the expected pattern for consumers

*Affected consumer files:*

• `jupyter/tensorflow/ubi9-python-3.12/pyproject.toml`
• `runtimes/tensorflow/ubi9-python-3.12/pyproject.toml`

Parent: #77