model.save not working

[wjdolan]

I am unable to save the model created in Transfer Learning part 3 due to the following error (I also couldn't find any suggestions to resolve this anywhere else):

KeyError: "Failed to add concrete function 'b'__inference_model_layer_call_fn_197900'' to object-based SavedModel as it captures tensor <tf.Tensor: shape=(), dtype=resource, value=> which is unsupported or not reachable from root. One reason could be that a stateful object or a variable that the function depends on is not assigned to an attribute of the serialized trackable object (see SaveTest.test_captures_unreachable_variable)."

Any suggestions?

[npsimid]

I have the same problem

[ashikshafi08]

Hey @wjdolan and @npsimid

It seems like one of the data augmentation layers caused this error, I tried to build the model without any data augmentation layer and it worked without any errors. Also, I just found that the reason came from the data_augmentation layer since the RandomFlip and RandomRotation and a few other layers are changed from layers.experimental.preprocessing.RandomFlip to layers.RandomFlip, but still modifying this throws an error.

Lemme raise this question in more brief at the TensorFlow forums till then try not to save the model which has data augmentation layers in them.

# Setup base model and freeze its layers (this will extract features)
base_model = tf.keras.applications.EfficientNetB0(include_top=False)
base_model.trainable = False

# Setup model architecture with trainable top layers
inputs = layers.Input(shape=(224, 224, 3), name="input_layer") # shape of input image

#x = data_augmentation(inputs) # augment images (only happens during training) # uncomment 

x = base_model(inputs, training=False) # put the base model in inference mode so we can use it to extract features without updating the weights
x = layers.GlobalAveragePooling2D(name="global_average_pooling")(x) # pool the outputs of the base model
outputs = layers.Dense(len(train_data_all_10_percent.class_names), activation="softmax", name="output_layer")(x) # same number of outputs as classes
model = tf.keras.Model(inputs, outputs)

Hope that helps, for now, we will update more about this as we get to know and also will post the forum link here asap.

Best,
Ashik.

[wjdolan]

Yes, thanks for figuring it out!

[mrdbourke]

Thank you for posting these. Will check them out ASAP and get back to you.

[UglyBaliCode]

I'm having the same issue (it seems to be related with Colab 2.7).
There is something on Stackoverflow about this issue ...
https://stackoverflow.com/questions/69955838/saving-model-on-tensorflow-2-7-0-with-data-augmentation-layer

temporary solution -> save it as h5

I hope there will be an easy way to implement solution ...