[exegete]

I’d agree on the training time but your serialized model should be small on disk since only the support vectors are needed for inference. At least with my experience that has been true.