Control token a pharse in list keyphrase using Text2Text

Hi everyone !
When I use model end2end model as (T5, MBart,…) for task (keyphrase abstractiion) after I use model.generate(…) to gen list keypharse separated by semi-colon
Examples: list_keypharse: language model;keyphrase generation;new pre-training objective;pre-training setup;

  • I only can control max_length of list_keypharse not control max_length of a keypharse(examples:new pre-training objective ( i want 2 token not 3 token)) in list_keypharse

How I can control and handle problems above? Thanks u everyoneThis text will be hidden