Loading of a model with a Bidirectional LSTM layer no longer working in TensorFlow 2.16.1 about tensorflow HOT 7 OPEN

Hi @Dobiasd ,

There is an open issue in Keras repo # 18913 where Keras Team looking into same and expect a fix ASAP.

Thanks!

from tensorflow.

@Dobiasd Could you try to use model.save('model.h5', save_format='tf') to save the model in TensorFlow SavedModel format as the default model.save might not capture all model components. Please have a look at this gist and let us know if it helps?
Thank you!

from tensorflow.

I tried:

from tensorflow.keras.layers import Input, LSTM, Bidirectional
from tensorflow.keras.models import Model, load_model

inputs = [Input(shape=(17, 4))]
layer = Bidirectional(LSTM(units=4), merge_mode='concat')(inputs[0])
outputs = [layer]
model = Model(inputs=inputs, outputs=outputs)

model.save('model.h5', save_format='tf')
load_model('model.h5')

model.save('model.h5', save_format='tf') raises a ValueError:

  File "h5py/h5d.pyx", line 137, in h5py.h5d.create
ValueError: Unable to synchronously create dataset (name already exists)

However, the model.h5 is written to the disk nonetheless.

But loading it (load_model('model.h5')) fails:

  File "h5py/h5o.pyx", line 189, in h5py.h5o.open
KeyError: "Unable to synchronously open object (object 'input_layer' doesn't exist)"

from tensorflow.

@Dobiasd Thank you for the response here.
@SuryanarayanaY I was able to replicate the issue reported here.
Thank you!

from tensorflow.

I am having the same issue. This and other issues have resulted in frustrating experiences when moving from previous versions of tf into tf 2.15 and TF 2.16, including 2.16.1.

from tensorflow.

Having same issue with tensorflow==2.16.1 and keras==3.2.1
Is downgrading really the only fix right now?

from tensorflow.

Hi @Dobiasd ,

Please note that Keras3 supports saving the model in .keras format only as It is now supporting multiple backends like Torch and Jax also.With .keras format this has been working fine in nightly and attached gist for reference.

If you want to save the model in tf format you might need to use model.export("saved_model") instead of model.save().

Please refer the migration guide here.

from tensorflow.