AI Accelerator - Pipelines 4.0.1 release notes

The knowledge base and preparer pipeline now support arbitrary Postgres schemas for source and destination tables/volumes. A bug prevented users for configuring explicit schemas via Postgres qualified identifiers (schema.name) when referencing source or destination tables/volumes in the create pipeline calls. This bug is now fixed.

A bug in the batch-processing code for knowledge base pipelines would, under certain circumstances, lead to inaccurate results during retrieval. This bug was introduced in the 4.0.0 release and is now fixed.

External models, running on the HCP Model Serving infrastructure, can now be listed and integrated into AIDB with new helper functions aidb.get_hcp_models() and aidb.create_hcp_model().

When using external models for embeddings (e.g. with the nim_embeddings model adapter), AIDB performs additional API calls in order to probe the model service for the response type. These calls are now skipped in most situations to reduce the overhead to a minimum when running embedding processing. AIDB will now use actual results to determine the response type.