Hi, I have a question about migrating from tensorflow 2.14 to 2.16.
Up until now I have been using tf.keras.callbacks.ModelCheckpoint to save models in the SavedModel format (by giving the callback a directory as the filepath ), as a result ClearML would "auto-magically" upload zip files with assets and variables subdirectories and fingerprint.pb and saved_model.pb files, which is exactly what I wanted.
Now in 2.16 it turns out that keras.callbacks.ModelCheckpoint no longer supports the SavedModel format. So I wrote my own callback that uses the tf.saved_model.save function. However, ClearML now uploads only the variables (as variables.zip ) for some reason, even though I inspect the path given to tf.saved_model.save and it indeed contains all expected files and subdirectories.

Any idea how to solve this?