So far the best outcome I've been able to gain is by manually creating the trieve data set and significantly increasing the splits per chunk.