I’m getting around to migrating to 1.x now finally. One thing I seem to be missing now is the ability to run a bot on the console and specify separately the NLU and Core models. This is quite inconvenient - consider the case where I have a lot of NLU training data and a complex NLU model that takes on the order of half an hour to train, and then I’m developing my Core stories. This makes for very slow iteration if I have to train a “combined NLU and Core” model every time I make a change to the stories, and want to check the outcome.
Am I missing something? Is it possible to use the “rasa shell” command and specify separately core and NLU models?
OK, by digging in the code a bit I’ve found out that the “rasa train” command will only retrain the NLU model if the NLU “fingerprint” (data+config) has changed, and same for core. Would have found that out if I had actually tried the command before complaining!
It would still be good to have the ability to specify the NLU and Core models separately when launching the rasa service. The reason being, I would like to train the NLU and Core models in parallel. With our models getting larger, and the scope of our bot expanding, training our bot is having a significant impact on developer productivity. Being able to train the models in parallel and selecting Core and NLU models when starting the service would save a lot of time.
Happy to see you found a solution