Also test how much batching speeds it up: https://stanfordnlp.github.io/stanza/pipeline.html#basic-example