Saving formats

[pedrobrs]

I used to save models with ModelCheckpoint callback and filename .h5.
Recently update tensorflow to version 2.13.

Now If save the model in h5 format, a warning is logged:
UserWarning: You are saving your model as an HDF5 file via model.save(). This file format is considered legacy. We recommend using instead the native Keras format, e.g. model.save('my_model.keras').

But If use keras, I hit this issue:
https://stackoverflow.com/questions/76701617/the-following-arguments-are-not-supported-with-the-native-keras-format-opti

and, if I use tf format , I get no logs for fit!
When checkpoint filename ends with .h5 I get a log like:
absl - 1106/1106 - 41s - loss: 0.0014 - accuracy: 0.9996 - precision: 0.9998 - recall: 0.9996 - auc: 1.0000 - val_loss: 0.0046 - val_accuracy: 0.9978 - val_precision: 0.9987 - val_recall: 0.9984 - val_auc: 1.0000 - lr: 0.0010 - 41s/epoch - 37ms/step

But with a filename .tf, that log is not generated.

I would like to use a non legacy format but without exceptions nor logs missing.
What can I do?

[SuryanarayanaY]

@pedrobrs ,

We have observed similar behaviour reported in TF repo as well recently and this issue persists from 2.13 version onwards. While saving model with .keras format and if model have a callback say for eg.

tf.keras.callbacks.ModelCheckpoint(
        "models/tmp_saved_model4" + "/{epoch}_model.keras", 
        save_best_only=True #making it to False or None works
)

It fails with similar error as reported in the attached SO link.

Similar issue was discussed in TF #61999. Please have a look into it.

As per my understanding there is issue with model.save('somename.keras', options=something) that causing the failure.

Could you please check and confirm for your case what are the options that are getting passed to model.save() ?

[SuryanarayanaY]

Hi @pedrobrs ,

I believe this issue should have fixed by this PR #18545. The relevant tensorflow issue #tensorflow/tensorflow#61999 also resolved with this PR.

Could you please check the issue with nightly and let us know if yet to resolve. Thanks

[rajveer43]

can I work on this?

[github-actions]

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

[rajveer43]

can I work on this?

[rickiepark]

Hi @SuryanarayanaY
Same issue is in TF 2.14, please check this [gist].(https://colab.research.google.com/drive/17Fbhpb5XJqRZURZm8wR9xyEM0wOf1Fdc)

[SuryanarayanaY]

Hi @rickiepark ,

Could you please check with nightly version as changes are not merged into nightly only? Also your colab gist not accessible.

[github-actions]

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

[github-actions]

This issue was closed because it has been inactive for 28 days. Please reopen if you'd like to work on this further.

[steveepreston]

this issue exists in TensorFlow 2.16.1

[ghsanti]

@steveepreston

this issue exists in TensorFlow 2.16.1

Can you share a minimal colab that reproduces the error?

[YMasterous]

This issue also exists in TensorFlow 2.17.0