Continuing to hack on ML ingest for search. On the sample I made yesterday it was a pretty significant improvement. The ingest keeps choking though due to tensor size even though I'm truncating the input significantly ... so much glue.