When I upload a model to Rasa X (using Helm chart deployment on GKE) with the latest version of Rasa X (0.33) and Rasa 2.0.2, talking to the bot does not work as it should:
The input is set equal to the intent (i.e. if I say hello the intent = hello but if I say bananas the intent = bananas (see picture below).
The bot keeps loading (…) (see picture below as well)
To be clear, you mean that the model works as expected locally?
If you’re seeing this behavior in Rasa X server mode but not running the model locally or in Rasa X local mode, it sounds suspiciously like a bug we’ve pinned down when using rasa open source >= 2.0.3. Are you sure that you are running 2.0.2?
If so, can you provide the logs of the rasa-production container? For some reason it seems that your NLU model is not getting loaded correctly, so a RegexInterpreter is being used in its place.
I think it’s the same original problem from Fabrice but I just followed the steps quick-install, nothing else, then at the first test on a simple bot … Rasa X doesn’t work properly!!