using model.fit instead of model.fit_generator

[filbertphang]

hello!

i am trying to build a classifier on top of the keras-bert model. however, when I attempt to run model.fit on the new model, I constantly get ValueError: An operation has `None` for gradient.

upon further testing, running model.fit on the loaded keras-bert model alone also gives this error. as such, may I clarify if fit_generator must be used for fine-tuningthe BERT model?

(my rationale for not using the original BERT code to fine-tune the model is because I am making use of keras shared layers as part of my classifier, which does not appear to be doable with tensorflow alone)

thank you so much!