-
Notifications
You must be signed in to change notification settings - Fork 54
Closed
Labels
Milestone
Description
Bug description
When running the 04 Retrieval Model Notebook if you try to save the two tower model after you run
model.evaluate(valid, return_dict=True, batch_size=1024)
model.save(os.path.join(DATA_FOLDER, "model"))
You get an error in the tf_utils.maybe_serialize_keras_objects function like this:
75 config[key] = {
76 k: tf.keras.utils.serialize_keras_object(v) for k, v in maybe_value.items()
77 }
78 elif isinstance(maybe_value, (list, tuple)):
---> 79 config[key] = [tf.keras.utils.serialize_keras_object(v) for v in maybe_value]
80 else:
81 config[key] = tf.keras.utils.serialize_keras_object(maybe_value)
AttributeError: 'NoneType' object has no attribute 'items'
Steps/Code to reproduce bug
- add the model save call after the model.evaluate call.
- run the notebook, trigger the save call.
Expected behavior
Successfully save model with all necessary artifacts.
Additional context
Would like to get this export working because it would allow me to use this dataset in the example for merlin systems.
Reactions are currently unavailable