TensorFlow.js Python Library

4.22.0 · active · verified Thu Apr 09

The `tensorflowjs` Python package provides utilities to convert trained Keras or TensorFlow SavedModel models into a format consumable by TensorFlow.js, enabling them to run directly in a web browser or Node.js environment. It facilitates the deployment of machine learning models on the web. The current version is 4.22.0, with releases typically aligned with major TensorFlow versions.

Warnings

breaking Compatibility with TensorFlow versions is crucial. Using `tensorflowjs` with a significantly older or newer `tensorflow` package can lead to conversion errors or unexpected behavior. Always check the official documentation for supported TensorFlow versions for your `tensorflowjs` package.
Fix: Ensure your `tensorflow` and `tensorflowjs` package versions are compatible. Typically, `tensorflowjs` versions are released to work with the latest stable `tensorflow` at that time. Update both packages or use versions known to be compatible.
gotcha Custom TensorFlow Operations or Keras Layers are not automatically converted to their JavaScript equivalents. If your model uses custom ops/layers, you will likely need to implement their logic manually in JavaScript for them to work in TensorFlow.js.
Fix: For models with custom components, consult the TensorFlow.js documentation on custom operations and layers. You may need to register custom ops in JavaScript or refactor your model to use standard layers.
deprecated While `.h5` (HDF5) Keras model files can often be converted, the recommended and most robust format for saving TensorFlow and Keras models is the TensorFlow SavedModel format (`model.save('path', save_format='tf')`). Conversion from SavedModel is generally more reliable.
Fix: Always save your TensorFlow/Keras models in the SavedModel format (`model.save('path', save_format='tf')`) and use `tfjs.converters.convert_tf_saved_model` or the `tensorflowjs_converter` CLI tool with the `tf_saved_model` input format.
gotcha Converting very large models can be resource-intensive (memory and CPU) and slow. Ensure your conversion environment has sufficient resources.
Fix: For extremely large models, consider model pruning, quantization (which `tensorflowjs_converter` supports), or splitting the model into smaller, manageable parts. Run the conversion on a machine with ample RAM and CPU.

Install

pip install tensorflowjs Install `tensorflowjs`

Imports

save_keras_model

from tensorflowjs.converters import save_keras_model

convert_tf_saved_model
```
from tensorflowjs.converters import convert_tf_saved_model
```
Used for converting TensorFlow SavedModel directories.

Quickstart

This quickstart demonstrates how to create a simple Keras model, train it, and then use `tensorflowjs.converters.save_keras_model` to convert it into the TensorFlow.js Layers format, ready for web deployment. The output is a directory containing `model.json` and weight files.

import tensorflow as tf
import tensorflowjs as tfjs
import os
import shutil

# Create a simple Keras model
model = tf.keras.Sequential([
    tf.keras.layers.Dense(units=1, input_shape=[1])
])
model.compile(optimizer='sgd', loss='mean_squared_error')

# Train the model (dummy data)
hs = model.fit([1, 2, 3, 4], [0, -1, -2, -3], epochs=1)

# Define output directory
output_dir = 'my_tfjs_model'
if os.path.exists(output_dir):
    shutil.rmtree(output_dir)

# Convert the Keras model to TensorFlow.js format
tfjs.converters.save_keras_model(model, output_dir)

print(f"Model converted and saved to: {output_dir}/")
print("You can now serve this model with `tensorflowjs_converter --input_format=tfjs_layers_model <path-to-model-json-file>` or directly load it in JavaScript.")

# Clean up (optional)
# shutil.rmtree(output_dir)

view raw JSON →