Multivariate Time Series RNN using categorical_column

I am trying to build a multivariate RNN for timeseries. Does anyone know how to use categorical_column_with_hash_bucket and categorical_column_with_vocabulary_list for some of the features when you want to use 20 past observations to predict the next 10? Most timeseries examples for TF use pandas/numpy, but with a lot of data it runs into memory issues