Adding TensorFlow Hub KerasLayer to Sequential Model Raises ValueError

March 25, 2024

When attempting to add a TensorFlow Hub KerasLayer to a Sequential model, you may encounter the following error:

Only instances of `keras.Layer` can be added to a Sequential model.

Only instances of `keras.Layer` can be added to a Sequential model. Received: <tensorflow_hub.keras_layer.KerasLayer object at 0x72492078c110> (of type <class 'tensorflow_hub.keras_layer.KerasLayer'>)

Cause

This error occurs because the isinstance(layer, Layer) check in Sequential.add returns False for hub.KerasLayer, even though it inherits from keras.layers.Layer. This is due to a change in the way Keras imports the TensorFlow backend in versions 2.16.0 and above.

Solution

To resolve this issue, you can use the following solution:

Install the tf_keras package using pip:

pip install tf_keras

In your code, use the following code to determine which version of Keras to import:

version_fn = getattr(tf.keras, "version", None)
if version_fn and version_fn().startswith("3."):
  import tf_keras as keras
else:
  keras = tf.keras

This code imports tf_keras if you are using Keras version 3 or higher, and imports tf.keras directly if you are using Keras version 2.

Example

Consider the following code snippet:

import tensorflow as tf
import tensorflow_hub as hub

model = tf.keras.Sequential()
layer = hub.KerasLayer("https://tfhub.dev/google/tf2-preview/nnlm-en-dim128/1")

# This line raises the error in TensorFlow 2.16.0 and above
model.add(layer)

To fix the error, you can use the following updated code:

import tensorflow as tf
import tensorflow_hub as hub

version_fn = getattr(tf.keras, "version", None)
if version_fn and version_fn().startswith("3."):
  import tf_keras as keras
else:
  keras = tf.keras

model = keras.Sequential()
layer = hub.KerasLayer("https://tfhub.dev/google/tf2-preview/nnlm-en-dim128/1")

model.add(layer)

With this modification, the code will run successfully without raising the error.