Rasa: 3.5.0 Release

Rasa 3.5.0 (2023-03-21)

Features

• #12053: Add a new required key (assistant_id) to config.yml to uniquely identify assistants in deployment. The assistant identifier is extracted from the model metadata and added to the metadata of all dialogue events. Re-training will be required to include the assistant id in the event metadata.

If the assistant identifier is missing from the config.yml or the default identifier value is not replaced, a random identifier is generated during each training.

An assistant running without an identifier will issue a warning that dialogue events without identifier metadata will be streamed to the event broker.

Improvements

• #11998: Add capability to send compressed body in HTTP request to action server. Use COMPRESS_ACTION_SERVER_REQUEST=True to turn the feature on.

Bugfixes

• #12136: Address potentially missing events with Pika consumer due to weak references on asynchronous tasks, as specifcied in Python official documentation.
• #12155: Sets a global seed for numpy, TensorFlow, keras, Python and CuDNN, to ensure consistent random number generation.

Improved Documentation

• #11893: Clarify in the docs, how rules are designed and how to use this behaviour to abort a rule

Miscellaneous internal changes

• #11924, #11926, #12006, #12022, #12092, #12135, #12137, #12140, #12154