Saving TensorFlow models for AI Explanations

This page explains how to save a TensorFlow model for use with AI Explanations, whether you're using TensorFlow 2.x or TensorFlow 1.15.

TensorFlow 2

If you're working with TensorFlow 2.x, use tf.saved_model.save to save your model.

A common option for optimizing saved TensorFlow models is for users to provide signatures. You can specify input signatures when saving your model. If you have only one input signature, AI Explanations automatically uses the default serving function for your explanations requests, following the default behavior of tf.saved_model.save. Learn more about specifying serving signatures in TensorFlow.

Multiple input signatures

If your model has more than one input signature, AI Explanations cannot automatically determine which signature definition to use when retrieving a prediction from your model. Therefore, you must specify which signature definition you want AI Explanations to use. When you save your model, specify the signature of your serving default function in a unique key, xai-model:

tf.saved_model.save(m, model_dir, signatures={
    'serving_default': serving_fn,
    'xai_model': my_signature_default_fn # Required for AI Explanations
    })

In this case, AI Explanations uses the model function signature you provided with the xai_model key to interact with your model and generate explanations. Use the exact string xai_model for the key. See this overview of Signature Defs for more background.

Preprocessing functions

If you use a preprocessing function, you also need to specify the signatures for your preprocessing function and your model function when you save your model. Use the xai_preprocess key to specify your preprocessing function:

tf.saved_model.save(m, model_dir, signatures={
    'serving_default': serving_fn,
    'xai_preprocess': preprocess_fn, # Required for AI Explanations
    'xai_model': model_fn # Required for AI Explanations
    })

In this case, AI Explanations uses your preprocessing function and your model function for your explanation requests. Make sure that the output of your preprocessing function matches the input that your model function expects.

Try the full TensorFlow 2 example notebooks:

TensorFlow 1.15

If you're using TensorFlow 1.15, do not use tf.saved_model.save. This function is not supported with AI Explanations when using TensorFlow 1, instead use tf.estimator.export_savedmodel in conjunction with an appropriate tf.estimator.export.ServingInputReceiver

Models built with Keras

If you build and train your model in Keras, you must convert your model to a TensorFlow Estimator, and then export it to a SavedModel. This section focuses on saving a model. For a full working example, see the example notebooks:

After you build, compile, train, and evaluate your Keras model, you have to:

  • Convert the Keras model to a TensorFlow Estimator, using tf.keras.estimator.model_to_estimator
  • Provide a serving input function, using tf.estimator.export.build_raw_serving_input_receiver_fn
  • Export the model as a SavedModel, using tf.estimator.export_saved_model.
# Build, compile, train, and evaluate your Keras model
model = tf.keras.Sequential(...)
model.compile(...)
model.fit(...)
model.predict(...)

## Convert your Keras model to an Estimator
keras_estimator = tf.keras.estimator.model_to_estimator(keras_model=model, model_dir='export')

## Define a serving input function appropriate for your model
def serving_input_receiver_fn():
  ...
  return tf.estimator.export.ServingInputReceiver(...)

## Export the SavedModel to Cloud Storage, using your serving input function
export_path = keras_estimator.export_saved_model(
  'gs://' + 'YOUR_BUCKET_NAME',
  serving_input_receiver_fn
).decode('utf-8')

print("Model exported to: ", export_path)

What's next