Is it necessary to finetune all data in libero?

[dstx123]

I noticed that pi0 performs well on Libero Spatial, Libero Object, Libero Goal, and Libero 10 datasets. However, the model I tried to pretrain on libero 90 does not work on these four datasets directly. Should I finetune these four datasets at the same time? Or can I finetune them separately and then evaluate them together?