Introduction

The error NameError: name 'load_model' is not defined appears when Keras model-loading APIs are used without the correct import. It usually happens after mixing TensorFlow and standalone Keras examples from different versions. The fix is straightforward, but version consistency is essential.

Core Sections

Use Correct Import for Current TensorFlow

In modern TensorFlow, import from tensorflow.keras.models.

1from tensorflow.keras.models import load_model
2
3model = load_model("saved_model.h5")
4print(model.summary())

This is the most common and stable import path in TensorFlow 2 projects.

Alternative Module-style Import

If you prefer module import style, reference through namespace.

from tensorflow import keras

model = keras.models.load_model("saved_model.h5")

Pick one style and use it consistently in the codebase.

Check Environment and Version Mismatch

If import still fails, verify interpreter and installed packages.

python -V
python -m pip show tensorflow
python -m pip show keras

Mixed environments often cause confusing import errors.

Distinguish SavedModel and H5 Paths

load_model supports both SavedModel directories and H5 files, but path mistakes are common.

1# H5 file
2model = load_model("model.h5")
3
4# SavedModel directory
5model = load_model("exported_model_dir")

Ensure the path exists and matches expected format.

Avoid Mixing Legacy keras and tf.keras

Using both keras and tensorflow.keras in one project can cause namespace and serialization issues. Standardize on tf.keras unless project constraints require otherwise.

Reproducible Loading Workflow

Keep save and load code aligned:

1from tensorflow.keras.models import Sequential
2from tensorflow.keras.layers import Dense
3
4m = Sequential([Dense(4, input_shape=(3,), activation="relu"), Dense(1)])
5m.save("demo_model.keras")
6loaded = load_model("demo_model.keras")

Round-trip tests catch incompatibilities early.

Notebook-specific Fixes

In notebooks, stale cells can cause NameError if imports were not rerun. Restart kernel and run from top when debugging inconsistent state.

Common Root Causes in Real Projects

This NameError often appears in three situations. First, copied snippets import from old standalone Keras. Second, notebooks execute cells out of order and skip imports. Third, virtual environments differ between terminal and notebook kernel. Diagnose systematically before changing model code.

import sys
print(sys.executable)

Confirm that executable path matches the environment where TensorFlow is installed.

Loading Custom Objects

If the import is correct but model loading still fails later, custom layers or losses may require custom_objects.

1from tensorflow.keras.models import load_model
2
3class MyLayer(tf.keras.layers.Layer):
4    def call(self, inputs):
5        return inputs
6
7model = load_model("model_with_custom.keras", custom_objects={"MyLayer": MyLayer})

This is a different error category, but teams often confuse it with missing import issues.

Reproducible Debug Checklist

Use a short checklist for fast triage.

• Verify import path
• Print TensorFlow version
• Print interpreter path
• Verify model file path
• Restart kernel and rerun notebook from top

A checklist reduces repeated trial-and-error and speeds onboarding for new contributors.

In shared repositories, enforce one import convention in linting rules and template notebooks. Standardizing this small detail prevents recurring NameError issues and improves team onboarding speed.

When model-loading code ships to production jobs, include startup checks that verify model files and framework versions before serving requests. Early validation fails fast and avoids partial-service failures that are harder to diagnose later.

Consistent tooling and templates make this class of import error much less frequent.

Add a lightweight startup health check in your application that attempts model load once and reports explicit import or path failures. This turns vague runtime NameErrors into actionable diagnostics for operators and developers.

Common Pitfalls

• Calling load_model without importing it.
• Mixing keras and tensorflow.keras APIs in one project.
• Running code in a different environment than the one where TensorFlow is installed.
• Loading wrong path type or typo in model file location.
• Debugging notebook state without restarting stale kernels.

Summary

• Import load_model from tensorflow.keras.models for TensorFlow 2 workflows.
• Keep save and load formats consistent.
• Verify active interpreter and installed package versions.
• Avoid mixing legacy and modern Keras namespaces.
• Use round-trip loading tests to keep model IO reliable.