Comment by c7b

Regarding the first, parallel requests to the same loaded model seem to work pretty well, I'm trying to find time to look more into it myself, but this may be something that might already be within reach for local models.