Recommended approach to handle long conversation stories that branches out at many conversation turns without using checkpoints

Hi there,

Building out a conversation design using Rasa OS 2.0 that has more than 30 conversation nodes for the longest conversation path and branches out at various conversation nodes for example, from start, at conversation node 5, 10, etc it branches to other conversation paths based on entity/slots/user-utterances.

I ran into issue where I was using checkpoints extensively and it got stuck while training exploding trackers.

Try to split the story up into shorter sections and use rules when possible. With rules, you can use wait_for_user_input: false at key branch points to reduce the length and number of stories.

1 Like