Is it possible to train the model in an incremental way from RasaX UI? I have noticed in the new version the navigation bar has the option to train model from scratch, upload model from files, or via API but there is no option to support incremental model training. Is this feature only supported in Rasa core but not in RasaX UI. If this is not currently supported, is there any plan to support it in the future? Please advise if there are alternate options to do incremental training from RasaX UI.
The main purpose of RasaX UI is to add new training examples to existing intents when they are not recognized properly. The incremental training from RasaX UI will definitely help to quickly add new training examples and generate & activate new model instead of waiting for hours to train model from scratch. This will really helpful once it is deployed into a production environment. We did upgrade the RasaX to the latest version 0.38.1 to explore incremental model training but it is very disappointing to know there is no support currently to do this. Please advise if there are any alternatives.
Hi @hari. You are correct. Today, incremental training is an experimental feature only supported in Rasa Open Source. It’s great to hear that you like it and would like it incorporated into Rasa X. We will decide based on feedback from users, such as yourself, about whether to make it generally available and add support for it in Rasa X.
Here is a workaround that can use today: annotate NLU data in Rasa X and then push your changes to your remote Git repo, which can automatically trigger a CI pipeline that uses incremental training to train a model much faster and a CD pipeline that then uploads that model to Rasa X (
curl -k -F "model=@my_model.tar.gz" "https://rasa-x-url/api/projects/default/models?api_token=XXXXXXXXXXXXXXX"). You could adjust the Rasa Train-Test Model GitHub Action to do this if you are using GitHub (or use it as inspiration if you are not)
Thank you @tyd for the response. As a workaround I was able to generate incremental modal though shell command from RasaX Docker container. The model is generated successfully and copied to the model mounted folder but it is not displaying in RasaX UI → Model tab. It’s not a surprise since I ran the command directly through shell so it didn’t update Tracker (PostgreSQL) database. Is there any shell command to update database? Hope to see incremental model training option in RasaX UI soon.
Maybe while inside the Rasa X container, you could do the curl command like @tyd said. But maybe save the output of the incremental training in a different directory first inside the Rasa X container, then do the curl command to upload.