Release notes for the Generative Models extension, version 2.1

New parameter completion type for OpenAI based models which replaces the previous logic for name-based completion type selection. This can lead to problems if chat-based model names did not start with “gpt” or “ft”. The explicit parameter now allows to select the type independent of the used names which leads to more robust behavior.
Documentation fix: The Qdrant database uses the parameter api_key in the dictionary connection instead of the parameter token. The documentation did not reflect this and showed token instead.
Bugfix: Confidences produced by ZeroShotClassification could have been delivered in the wrong order
Bugfix: Fixed a bug in collection creation for Qdrant which prevented insertion for all Qdrant versions larger than 1.7.4