keras.models.load_model ValueError: Unknown layer: Functional

[GO]

Hello,

I'm trying to load my model on an EC2 environment but i have the following issue:

ValueError: Unknown layer: Functional

i'm using the following command in python:

from tensorflow import keras

model = keras.models.load_model('/home/ubuntu/full_model.h5')

On Google colab it works perfectly but not on EC2.

Does anyone have this issue? I don't find anything on other forums.

[muhammad harris]

Hi. I think the model was trained on the recent tensorflow 2.3 version that names all nodes created with functional api with a 'functional' prefix. You may want to upgrade your EC2 tendorflow to 2.3 then try loading the model.

--
You received this message because you are subscribed to the Google Groups "TensorFlow Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to developers+...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/developers/1309874b-7dcb-4cbd-88c1-d526cd6d8d79o%40tensorflow.org.

[Aditya Srivastava]

Did you check the versions? 

Python, Tensorflow, try to keep them the same.

I'd suggest making a venv

Make a requirements.txt from Colab and use it to install dependencies on EC2's venv.

[Ilya Persky]

Looks like the backward compatibility was broken recently:
https://github.com/tensorflow/tensorflow/issues/41929

Check this out for the workaround:
https://stackoverflow.com/questions/63068639/valueerror-unknown-layer-functional

> To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/developers/CAJMT-dEDyFCE2UrfgmD-nKdqdXfzMw%2BPhDUpk%2BsGVeZHjCGoaw%40mail.gmail.com.



--
Thank you,
Ilya.

[GO]

Thank you all! Yes I was using an old version of tf... :)

[Francois Chollet]

This appears to be a forward compatibility issue: you cannot load a model saved with TF 2.2+ using an earlier version of TF. In general, forward compatibility is never guaranteed, but backwards compatibility is.

However you should be able to load earlier models using a more recent TF version.

The fix here should be for you to upgrade the TF version of the VM where you load the model to match or exceed the version number of the VM where you saved the model. (i.e. make sure to use TF 2.2 or TF 2.3).

Please let us know if that works for you.

Best,

Francois

--

You received this message because you are subscribed to the Google Groups "TensorFlow Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to developers+...@tensorflow.org.

To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/developers/819629c3-d161-4c1f-bce6-6108997b8605o%40tensorflow.org.

[GO]

Hello Francois,

Yes thank you for the precision. Now it works perfectly following the upgrade of tf.

Actually, I was not using a good instance of EC2... I needed to choose another one for ML.

Thanks everyone again :)

Le vendredi 7 août 2020 13:22:33 UTC+2, GO a écrit :