NumPy 2: 'A module that was compiled using NumPy 1.x cannot be run in NumPy 2.0.0 as it may crash' / 'numpy.core.multiarray failed to import' after upgrading numpy. How do I fix it?

Cause: NumPy 2.0 changed its binary interface (ABI). Compiled extension modules (in packages such as pandas, scipy, scikit-learn, matplotlib, OpenCV, torch add-ons, or your own C/Cython code) that were built against NumPy 1.x use the old interface and cannot be loaded with NumPy 2. The error appears when an older compiled package is installed next to a freshly installed numpy 2.x, which pip does by default when nothing caps it. Wheels built against NumPy 2.x work with both 1.x and 2.x at runtime, so the fix is to get such builds, or to stay on 1.x. Fix (best first): 1. Upgrade the packages named in the traceback (and the ones that depend on numpy) to releases that support NumPy 2: pip install --upgrade <package>. Check each project's release notes for NumPy 2 support. 2. If a needed package has no NumPy 2 build yet, pin numpy below 2 in that environment: pip install "numpy<2", and record it in requirements.txt or pyproject.toml. Treat this as temporary. 3. Use a clean virtual environment rather than patching a mixed one, and re-resolve all dependencies together (pip install -r requirements.txt, or uv/poetry lock) so no old binary wheel is left behind. 4. If you maintain a package with compiled code: build against NumPy 2.x (build requirement numpy>=2.0 or newer) and keep a looser runtime requirement such as numpy>=1.23.5 if you still support 1.x; NumPy documents that this produces wheels for both. Use Cython 3, and consider ruff's NPY201 rule to find removed NumPy APIs in Python code. 5. Do not pin NumPy 1.x permanently, and do not install packages with --no-deps just to bypass the version conflict. Verify: python -c "import numpy, <package>; print(numpy.__version__)" imports cleanly with no ABI warning, and pip check reports no conflicts. Sources: https://numpy.org/doc/stable/numpy_2_0_migration_guide.html https://numpy.org/doc/stable/dev/depending_on_numpy.html https://numpy.org/doc/stable/release/2.0.0-notes.html