How to export a model for TensorFlow.js

This post shows how to save a “hello world” model in TensorFlow (Python), export it for TensorFlow.js, then run it in the browser. The model multiplies its input by 5, and you can see this exciting behavior here:

Here’s some Python that saves a model in the SavedModel format. The model just multiplies its input by 5.

import tensorflow as tf

class HelloModule(tf.Module):
  def __init__(self):
    super(HelloModule, self).__init__()
  @tf.function(input_signature=[tf.TensorSpec(shape=[None], dtype=tf.float32)])
  def __call__(self, inputs):
    return inputs * 5.

module = HelloModule(), './hello_model/saved_model')

After running this, you can see various saved files:

$ find hello_model

You then use tensorflowjs_converter to convert the SavedModel format to the (undocumented) TensorFlow.js “Graph Model” format:

$ pip3 install tensorflowjs
$ tensorflowjs_converter \
  --input_format=tf_saved_model \
  --output_format=tfjs_graph_model \
  hello_model/saved_model hello_model/web_model

Writing weight file hello_model/web_model/model.json...

This has created a model.json which refers to some binary files containing the weights:

$ find hello_model/web_model

Finally, here’s a webpage that uses TensorFlow.js to load the model, then runs the model on input from the user:

<script src=""></script>
<input type="number" id="modelInput" />
<span id="modelOutput"></span>
  const modelInputEl = document.getElementById("modelInput");
  const modelOutputEl = document.getElementById("modelOutput");
  (async () => {
    const model = await tf.loadGraphModel('./hello_model/web_model/model.json');
    modelInputEl.addEventListener("input", () => {
      const inputFloat = parseFloat(modelInputEl.value);
      tf.tidy(() => {
        modelOutputEl.innerText = model.predict(tf.tensor([inputFloat])).arraySync()[0];
I just released Vidrio, a free app for macOS and Windows to make your screen-sharing awesomely holographic. Vidrio shows your webcam video on your screen, just like a mirror. Then you just share or record your screen with Zoom, QuickTime, or any other app. Vidrio makes your presentations effortlessly engaging, showing your gestures, gazes, and expressions. #1 on Product Hunt. Available for macOS and Windows.

With Vidrio

With generic competitor

More by Jim

Tagged #tensorflow, #programming, #ml. All content copyright James Fisher 2021. This post is not associated with my employer. Found an error? Edit this page.