They’re fairly useful as constrain domain classifiers due to the low memory requirements means you can stuff a lot of them into a less expensive GPU farm and ge

They’re fairly useful as constrain domain classifiers due to the low memory requirements means you can stuff a lot of them into a less expensive GPU farm and get really decent throughout with pretty good results over all. At least that’s my experience. I wouldn’t bother using a tiny model for coding - but the world is full of abductive reasoning tasks that don’t involve coding.