You can speed up the map function by setting batched=True to process multiple elements of the dataset at once:

tokenized_billsum = billsum.map(preprocess_function, batched=True)

Now create a batch of examples using [DataCollatorForSeq2Seq].