0.3.3

more robust operation for MM/ML tags

update the call_mods model (v1 to v2), use shared params for both rnn and attention

update aggregate model (v2 to v2p)