Also a question on evaluations. I'm using Vertex AI's batch processing directly with some advanced features, this writes a file to GCS. So my workflow is batch process to caption a list of images followed by batch process with llm as judge on the output captions. I have multiple llm-as-judge evaluations I run. What is the best practice for importing these into Arize? Is there a way to incrementally add "columns" to a dataset or do I need to wait until all processing is done to upload? What about when I tune my llm-as-judge prompt? Still trying to grok some of the paradigms and integrate with my current processes.
