I'm currently working on a project where we're building a data pipeline. We have spark setup and have generated models. Sadly, loading the model in Spark and querying it isn't fast enough for us. What's the most straightforward way of exporting a model and loading it into memory on a local server? I've researched PMML and some libraries and that appears to be one path.
If PMML supports the model you're trying to express, it's probably the best option. Spark has some partial support for exporting as PMML. Look into JPMML + OpenScoring to score the PMML in an embedded instance.
There are loads of tools in this space that implement something like this, but they're typically entwined with their own formats or model management system.