not something i need for what i'm doing, so no. but i'd like to sometime. never tried doing it myself

It helps to have some decent hardware that you're training on. And keep your expectations relatively low. The small models are very different than the frontier models. But if you have a narrow, low-stakes task, it's kind of incredible to be able to get LLM responses locally free and nearly instant.
For instance, maybe you're are creating a game and you have a structured representation of the current scene. But the structured representation is boring for a player to see. So you could have a small model like this take that structured data and describe it more illustratively. Or the other way around: you want to take mushy user input and map it onto a small set of structured commands. So basically a suggestion feature for a known set of capabilities. Small models are good at things like that. You could imagine having an entire piece of software where the user never has to understand what commands are available to them. They just describe what they want to do and a local model interprets the intention into actions.

Qwen 3.5 2B for instance. how? ask another ai, they'll happily do this for you. if you have a narrow, trainable task/skill you can often get something competent at under a couple gigs

i've never met a person who is "satisfied" - in the *ego* sense - with acclaim accruing to a pseudonym and not their meatspace identity. most people seem to feel "unseen" unless the seen is attached to their body. for good, deep, evolutionary reasons, too.

what is amethyst doing for you? it's great for me

what were you doing with it..?

queued, thanks