Developer Setup & Documentation Issue (Frontend Related)
ISSUE2: The TFLite tests (e.g., test_frontend_tflite.py) fundamentally require tensorflow==2.9 to avoid _ARRAY_API core dumps via Numpy 2.x and to maintain API compatibility
python=3.10
i ran an upgrade to the pip defaults (tensorflow>=2.16.0 and numpy>=2.0.0)
and ran the frontend tflite tests again pytest tests/python/relax/test_frontend_tflite.py -v
got: Segmentation fault (core dumped)
Testing the frontend with the new TensorFlow and NumPy version: tensorflow>=2.16.0 numpy>=2.0.0

After downgrading to stable versions: tensorflow==2.9.0 numpy==1.26.4
ISSUE2: everything runs correctly if the above issue was resolved, BUT if you used python>=3.11 you'll face errors again
python=3.11.15
- Although in The Install from Source docs explicitly instruct users to set up their environment with
python>=3.10.
- old TF/Numpy refuses to install on Python 3.11
TensorFlow 2.9 has no pre-compiled binaries for Python 3.11+. If a user follows the official documentation and installs Python 3.11, they physically cannot install the correct TensorFlow version, forcing a pip upgrade chain that ultimately causes C++ ABI crashes and test failures.
Developer Setup & Documentation Issue (Frontend Related)
ISSUE2: The TFLite tests (e.g., test_frontend_tflite.py) fundamentally require tensorflow==2.9 to avoid _ARRAY_API core dumps via Numpy 2.x and to maintain API compatibility
python=3.10i ran an upgrade to the pip defaults (
tensorflow>=2.16.0andnumpy>=2.0.0)and ran the frontend tflite tests again
pytest tests/python/relax/test_frontend_tflite.py -vgot: Segmentation fault (core dumped)
Testing the frontend with the new TensorFlow and NumPy version:

tensorflow>=2.16.0numpy>=2.0.0After downgrading to stable versions:
tensorflow==2.9.0numpy==1.26.4ISSUE2: everything runs correctly if the above issue was resolved, BUT if you used python>=3.11 you'll face errors again
python=3.11.15python>=3.10.TensorFlow 2.9 has no pre-compiled binaries for Python 3.11+. If a user follows the official documentation and installs Python 3.11, they physically cannot install the correct TensorFlow version, forcing a pip upgrade chain that ultimately causes C++ ABI crashes and test failures.