Felt like there used to be a thriving fine-tuning community a few years back - and then once we started getting models that were smart enough and generalist enough (i.e. post Llama-3-8b era) things kind of dropped off a little. Less need for fine-tunes when prompt-tweaking can get you most of the way if your base is smart enough I suppose? I do miss it - felt like more or less every week I'd open this sub to find some new weird and wonderful thing going on with home brewed models trained on Unsl...