Cant use core with seprate NLU without training NLU and core

So for my use case, I already have an NLU server up and running remotely.
I want rasa core to use that server which I’ve defined in the endpoints file, but if I just train a core model separately and run it in shell I doens’t call the remote NLU server at all and instead just uses ‘RegexInterpreter’.
But if add even the smallest fraction of NLU training data and train both core and NLU as one, then when running shell it correctly calls my remote NLU.
Is there a way I can only use that remote NLU and not have to train them both together, am I missing something?

Hey @JamieDF can I ask what version of Rasa you’re using? From a quick check of the code, it should only return the RegexInterpreter if the endpoint config is empty

Hey @akelad, I’m on ‘1.3.9’. Hmm yeah the endpoint config for sure exists as I have an actions endpoint in there that works

@JamieDF do you think you could upgrade to a newer version? e.g. 1.6.1? Because the way the interpreter is loaded has changed since version 1.3.9, so the issue may have been fixed already

@akelad Okay sure I’ll give it a shot

@akelad Just tested it on 1.6.1 and it works as it should.
Thank you!
and sorry for bothering you, I’ll update first next time :grimacing:

no worries, glad it’s resolved!