Fix ImportError: cannot import name 'abs'

ImportError: cannot import name 'abs' is confusing because abs() is a builtin you never import. The error isn't about your code importing abs — it's a package failing to load itself, and one of its internal imports (from ... import abs) fails along the way. The real question is always: which package is broken, and why.

Read the traceback first

The line that matters is the one naming the package whose import failed — often TensorFlow or one of its submodules. That tells you where to look. The message "cannot import name 'abs'" is a symptom of that package being shadowed, half-installed, or caught in a circular import.

The causes, and how to tell them apart

1. A local file shadows the real package (most common)

If you have a file named tensorflow.py, math_ops.py, numbers.py, or similar in your working directory — or a leftover __pycache__/.pyc — Python imports your file instead of the installed package, and the expected internals aren't there. Confirm it:

python -c "import tensorflow; print(tensorflow.__file__)"

If that path is in your project directory instead of site-packages, you've found it.

2. A broken or version-mismatched install (the classic TensorFlow case)

A partial install, or a TensorFlow/protobuf version mismatch, leaves the package unable to complete its own imports. Reinstall cleanly, ideally in a fresh environment:

python -m venv .venv && source .venv/bin/activate
pip uninstall -y tensorflow protobuf
pip install "tensorflow==2.15.*"   # a known-good, matched release

Using a virtualenv avoids the leftover-files problem that causes half of these errors.

3. A circular import

Module A imports B at the top, and B imports a name from A before A finishes loading. A is only partially initialized, so the name is missing. Break the cycle by importing inside the function that needs it, or by moving the shared symbol into a third module both can import.

4. A user-defined name shadowing the builtin

If your own module has abs = something or a file literally named abs.py, a later from mymodule import abs can pick up the wrong thing. Rename it — shadowing builtins is worth avoiding regardless.

Diagnose it in order

1. Look at the traceback — which package's import failed?
2. Check for shadowing: print(pkg.__file__); look for stray .py files matching package names in your cwd; delete __pycache__.
3. Check the install: pip show <package>; reinstall in a clean venv if the version looks off.
4. Check for cycles: if it's your own package, look for two modules importing each other at load time.

Prevent it

• Never name a file after a package you import (tensorflow.py, email.py, random.py).
• Work inside a virtualenv so installs are clean and reproducible.
• Pin compatible versions for frameworks with tight internal coupling (TensorFlow + protobuf, for example).
• Clear caches after refactors: stale .pyc files cause "impossible" imports.

Once you identify which package is actually failing to import, the fix is almost always: remove the shadowing file, reinstall cleanly, or break the circular import.

Sources

• TensorFlow API docs