Support of many projects using single agent with Rasa 1.0

Hi All,

We are doing the migration of two agents dialgoflow to RASA.

With rasa nlu 0.15 we was able to train multiple nlu models for the same agent (using --project option)

With rasa 1.0 I didn’t find how to do that and maybe it’s not supported in the new version of rasa.

*Before: python -m rasa_nlu.train --config config_rasa.yml --data training_data_toto.json --project toto

*Now: rasa train nlu -c config_rasa.yml -u training_data_toto.json (we can’t specify the project)

Could someone help us to know how to manage many models with the same agent.

Thanks, Omar

Hi @Omar, great that you are looking into Rasa 1.0!

Yes, in Rasa 1.0 we removed projects. We wrote a post a couple of weeks ago on that decision: Removing projects for Rasa NLU server. Feel free to comment with your feedback.

With Rasa 1.0 an agent can just have one model loaded at a time. If you are running a Rasa server, you can replace the loaded model of an agent by calling PUT /model (see https://rasa.com/docs/rasa/api/http-api/#tag/Model/paths/~1model/put). So you can either load the desired model before answering a request or you can run two Rasa servers - one for each model. As this solution might not be optimal, we are still collecting feedback. So, please share your thoughts :slight_smile:

1 Like