If you are actively indexing, those 2 cpus are going to be saturated because the base model is pretty large (110M parameters) and inference on CPU is very intensive.