⚠️ Pin absl-py <2.0 for TensorFlow 2.14 compatibility in ROCm notebook images

[coderabbitai]

Problem

TensorFlow 2.14 (including tensorflow-rocm==2.14.0.600) is incompatible with absl-py 2.x due to the removal of absl.logging APIs. This causes runtime import errors and prevents the image from starting.

Affected files

• jupyter/rocm/tensorflow/ubi9-python-3.11/requirements.txt
• (Potentially other TensorFlow 2.14-based images in the repository)

Impact

If absl-py >=2.0 is installed, TensorFlow 2.14 will fail at runtime with import errors (absl.logging.* not found). This breaks all notebook images using this combination.

Solution

• Pin absl-py to <2.0 in all affected requirements.txt files, OR
• Upgrade TensorFlow to a version compatible with absl-py 2.x (not possible for ROCm as of July 2025).

Acceptance criteria

• All TensorFlow 2.14 (including ROCm) notebook images pin absl-py to <2.0 in requirements.txt.
• CI and runtime tests pass without absl.logging import errors.
• This issue is cross-referenced in future PRs that address the pin or upgrade.

References

• PR #1325
• Relevant discussion
• Previous discussion

Requested by @jiridanek in PR #1325 review.