Find the original model card
here.
Making these was a lesson in pain and humility. It has been over two months since the day I decided "hm today i will learn how to make exl2 quants" <- (clueless). First my conda env
stopped working (for some reason), then it stopped recognizing venvs when I tried using those, then the universe decided to screw up the one venv I had working somehow (I can only assume
it was a cosmic bitflip or something because it literally stopped working overnight) and making these four quants alone took over an hour on my hardware, in which time I could probably have
made an entire set of GGUFs (plus a full set of i-quants) for three different models. Then uploading these was such a pain because huggingface-cli might as well be arcane magic since their
documentation doesn't really tell you how to actually use it or what exactly will happen when you run this method. I haven't even tested any of these quants because tabbyapi, to this day,
simply will not work. Torch keeps bugging me about running out of VRAM, even when trying to load 3bs. I have basically tried everything to try and get tabbyapi to run. It simply will not.
Suggest more models in the community tab and I might have a crack at exl2'ing them.
The model is quite nice though, it is quite useful for my usecases of synthetic variant generation if you're into that sort of thing.
