Tensorflow sparse ops require sorted indices

Question

I\'m trying to fine tuning Bert for document classification.

I started by tokenizing the documents to generate the input_ids, attention_mask