Great question! We have a built-in integration test for 1 million images [1] and have benchmarked EVA on large unstructured datasets for our ACM SIGMOD paper [2].
Coming up with an open benchmark for AI-Relational database systems is essential [3]. Please share your thoughts on any ideas for benchmarking :)
[1] https://github.com/georgia-tech-db/eva/blob/e123820c79b902d5...