Smaller version?
I have little knowledge of these things, but is there a way to train a student model, or, smaller versions of this? ive grown very curious about nemotron sunfall but by PC simply cannot handle 70b parameter models. the best i can do reasonably are 12b. and ive seen other models that have 70b, 30b, 22b etc. though i have no idea if that is an unreasonable request. id love to see the same of quartetanimoi
There are smaller sunfall variants out there, although they're not specifically Nemotron. I know there's another smaller Nemotron that Nvidia distilled from the bigger one. I'm not sure if it would be possible to do the training on that, but I could look into it. I'm not sure what quartetanimoi is.
He means https://huggingface.co/alchemonaut/QuartetAnemoi-70B-t0.0001 which is miqu-based.
Incidentally, that is the model I used before switching to this one. Never touched it again since then. But for a long time, nothing even came neart to it in terms of instruction following, especially if your instructions were somewhat nonstandard. Alas, miqu doesn't have 22b etc. variants, so there is no good way to achieve your goals, Utochi.