[Fixed] tensorflow.python.framework.ops Has No Attribute _tensorlike

The error AttributeError: module 'tensorflow.python.framework.ops' has no attribute '_tensorlike' usually means some code is depending on a private TensorFlow internal attribute that is not available in your installed TensorFlow/Keras combination. It often appears after upgrading TensorFlow, mixing standalone keras with tf.keras, or running an older third-party package against a newer TensorFlow release.

The most reliable fix is not to patch TensorFlow files manually. First identify your installed versions, then either upgrade the package that triggers the error or use a compatible TensorFlow/Keras environment for that project.

Contents

Quick diagnosis

Run this in the same virtual environment, notebook kernel, or runtime where the error appears:

import sys
import tensorflow as tf

print(sys.version)
print("TensorFlow:", tf.__version__)
print("tf.keras:", tf.keras.__version__ if hasattr(tf.keras, "__version__") else "bundled")

try:
    import keras
    print("standalone keras:", keras.__version__)
except ImportError:
    print("standalone keras: not installed")

If the script shows both tensorflow and standalone keras, check whether your code or dependency expects Keras 2, Keras 3, or tf.keras. TensorFlow 2.16 and newer made Keras 3 the default Keras version, while legacy Keras 2 support moved to the tf-keras package.

Fix 1: Update TensorFlow, Keras, and the failing package

For most users, start by updating the packages in a clean environment:

python -m pip install --upgrade pip
python -m pip install --upgrade tensorflow keras

Then update the package that imports TensorFlow internals. For example, if the traceback points to mtcnn, bert4keras, tensorlayer, or another wrapper library, upgrade that package too:

python -m pip install --upgrade mtcnn
python -m pip install --upgrade tensorlayer

Use the package name shown in your traceback. After upgrading, restart the notebook kernel or Python process so the old modules are no longer loaded.

Fix 2: Stop mixing incompatible Keras imports

Choose one Keras path for the project. Modern TensorFlow examples commonly use tf.keras or Keras 3 imports, but old code may mix both styles:

# Avoid mixing these styles in the same project unless the dependency docs require it.
from tensorflow import keras
import keras

If your project is current and supports Keras 3, prefer Keras 3 style imports:

import keras
from keras import layers

If the project is tied to TensorFlow’s Keras API, keep the imports consistent:

import tensorflow as tf

model = tf.keras.Sequential([
    tf.keras.layers.Dense(16, activation="relu"),
    tf.keras.layers.Dense(1),
])

Fix 3: Use legacy Keras only when a dependency requires it

Some older libraries still expect Keras 2 behavior. TensorFlow’s 2.16 release notes explain that Keras 2 remains available through the tf-keras package. Use this only for projects that have not migrated yet:

python -m pip install "tf-keras~=2.16"

If the dependency specifically says to use legacy tf.keras, set the environment variable before importing TensorFlow:

import os
os.environ["TF_USE_LEGACY_KERAS"] = "1"

import tensorflow as tf

This setting affects the Python process, so use a separate virtual environment for older projects. Do not set it globally across unrelated TensorFlow work.

Fix 4: Replace private TensorFlow checks in your own code

If the failing line is in your own code and it references tensorflow.python.framework.ops._tensorlike or _TensorLike, replace that private check with public TensorFlow APIs. For example:

import tensorflow as tf

def ensure_tensor(value):
    if not tf.is_tensor(value):
        value = tf.convert_to_tensor(value)
    return value

tf.is_tensor() and tf.convert_to_tensor() are public APIs. Private paths under tensorflow.python can change between releases, so code that imports from those paths is more likely to break.

Fix 5: Pin versions only as a temporary workaround

If a library has not released a compatible update, pinning versions can unblock a project temporarily:

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
python -m pip install --upgrade pip
python -m pip install "tensorflow==2.15.*" "keras<3"

Use this as a project-specific workaround, not a permanent system-wide fix. Record the reason in requirements.txt and revisit the pin when the package supports newer TensorFlow/Keras versions.

When the error appears in MTCNN, bert4keras, or TensorLayer

For third-party packages, read the traceback from bottom to top and identify the first non-TensorFlow package path. That package is usually the one importing a private TensorFlow symbol. The fix is normally one of these:

Upgrade the package to a version that supports your TensorFlow/Keras release.
Install the package's documented TensorFlow version in a clean virtual environment.
Use legacy tf-keras only if the package has not migrated to Keras 3.
Open an issue or patch the dependency if it still imports tensorflow.python.framework.ops._tensorlike.

For general Python AttributeError debugging patterns, see our guide to Python AttributeError examples.

Official references

Conclusion

The tensorflow.python.framework.ops _tensorlike error is usually a compatibility problem, not a missing line you should add to TensorFlow. Update the failing dependency first, keep Keras imports consistent, and use legacy tf-keras only when an older project requires it. If your own code imports TensorFlow internals, replace those imports with public APIs such as tf.is_tensor() and tf.convert_to_tensor().