Rasa: 3.2.0 Release

• #10989: [NLU training data](nlu-training-data.mdx) in JSON format is deprecated and will be removed in Rasa Open Source 4.0. Please use rasa data convert nlu -f yaml --data <path to NLU data> to convert your NLU JSON data to YAML format before support for NLU JSON data is removed.

Improvements

• #10696: Make TrackerStore interface methods asynchronous and supply an AwaitableTrackerstore wrapper for custom tracker stores which do not implement the methods as asynchronous.
• #10897: Added flag use_gpu to TEDPolicy and UnexpecTEDIntentPolicy that can be used to enable training on CPU even when a GPU is available.
• #11102: Add --endpoints command line parameter to rasa train parser.

Bugfixes

• #11129: The azure botframework channel now validates the incoming JSON Web Tokens (including signature).

Previously, JWTs were not validated at all. - #9386: rasa shell now outputs custom json unicode characters instead of \uxxxx codes

Improved Documentation

• #11130: Clarify aspects of the API spec
  • GET /status endpoint: Correct response schema for model_id - a string, not an object.
  • GET /conversations/{conversation_id}/tracker: Describe each of the enum options for include_events query parameter
  • POST & PUT /conversations/{conversation_id}/tracker/eventss: Events schema added for each event type
  • GET /conversations/{conversation_id}/story: Clarified the all_sessions query parameter and default behaviour.
  • POST /model/test/intents: Remove JSON payload option since it is not supported
  • POST /model/parse: Explain what emulation_mode is and how it affects response results

Miscellaneous internal changes

• #11088, #11123, #9093, #9099