MLFlow integration is great, eval stores... all the MLOps goodness. Partitioning is also fantastic. However, these two do not play well. We make use of partitioned models quite a bit because we love them. The unfortunate thing is we cannot train partitioned models with MLFlow and starting from 11.2 we cannot use eval stores with partitioned models (before they were supported but not accurate). We'd love to see more support for partitioned models across the MLOps and MLFlow integrations
