Model weights cannot be saved

[gcelano]

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

No

Source

source

TensorFlow version

2.16.1 and inferior

Custom code

Yes

OS platform and distribution

Debian 12 Bookworm

Mobile device

No response

Python version

No response

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

After model.save_weights is used, if a new process is started and model.load_weights is used, the results produced by the model are poor (very different from the ones at training time). The code can be found at https://git.informatik.uni-leipzig.de/celano/dithrax_test (with instructions).

(Since it is not clear if it is a keras or tensorflow problem, an issue has also been opened at keras-team/keras#19693 (comment))

Standalone code to reproduce the issue

https://git.informatik.uni-leipzig.de/celano/dithrax_test

Relevant log output

No response

[SuryanarayanaY]

Hi @gcelano,

In Keras3, In order to save weights using model.save_weights method you need to add .weights.h5 extension to the path.

model.save_weights('dir/my_model.weights.h5')

Please check the doc source for same.

[testhowtest]

These issues might share common themes or underlying causes related to saving and loading model weights in TensorFlow. Exploring them further could provide insights into resolving the "Model weights cannot be saved" issue. #65279 , #62483 , #61859 , #61279 , #60064 , #59287

[gcelano]

@SuryanarayanaY , in my code at link above I use ".weights.h5", and this has no impact on the problem. @testhowtest, is there a provisional workaround? Without saving, there is not much use in using TF

[tilakrayal]

@gcelano,
Looks like this is the Keras issue. It's better to track in the Keras repo which was raised for the quick resolution. Thank you!

[gcelano]

An issue has been opened here keras-team/keras#19693

[tilakrayal]

@gcelano,
Could you please feel free to move this issue to the closed status, as the same issue can be tracked in the Keras repo issue. Thank you!

[gcelano]

Closed because the issue is here: keras-team/keras#19693

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No