Keras 3 / TF SavedModel export issues about keras

[fchollet]

We don't recommend using tf.saved_model.save() for writing a SavedModel from a Keras model.

The recommended way is model.export("path"). If you call it on your model, you will see:

Saved artifact at 'path'. The following endpoints are available:

* Endpoint 'serve'
  args_0 (POSITIONAL_ONLY): TensorSpec(shape=(None, 10), dtype=tf.int32, name='x')
Output Type:
  TensorSpec(shape=(None, 2), dtype=tf.float32, name=None)

The input dtype and the input name are preserved. The output tensor name is not the name of the output layer, however. If you want to maintain that you will have to insert a tf.identity(..., name=...) op at the end of your model.

If you need more granular control over the endpoints and their signatures, you should use keras.export.ExportArchive.

If you want the behavior of tf.saved_model.save() to match model.export(), then you should file an issue with TensorFlow (we have no control over that function).

[wamuir]

Thanks! That helps a lot.

We don't recommend using tf.saved_model.save() for writing a SavedModel from a Keras model.

Here's what led me down that path:

keras/keras/saving/saving_api.py

Lines 110 to 111 in 5dd801a

	"Use `tf.saved_model.save()` if you want to export a SavedModel "
	"for use with TFLite/TFServing/etc. "

[fchollet]

Good catch, we should update that comment. That path does work most of the time but it isn't nearly as reliable as the one we actually control (model.export()).