"Auto class" functionality does not work. keras_nlp.models.Backbone("hf://google-bert/bert-base-uncased") will error.
Config overrides do not work. keras_nlp.models.BertBackbone("hf://google-bert/bert-base-uncased", dtype="bfloat16") will not silently ignore the dtype setting.
I also think there's a few places config overrides do not work outside of the transformers converts. E.g. instantiating a full task or preprocessor object.
keras_nlp.models.Backbone("hf://google-bert/bert-base-uncased")
will error.keras_nlp.models.BertBackbone("hf://google-bert/bert-base-uncased", dtype="bfloat16")
will not silently ignore the dtype setting.