Blessed be Sao who came down from the heavens to give us another 14b model. There's a 32b link but it's not up yet unfortunately. I'm sure when I post this it'll be up as a gotcha to me personally by God.
Anyway, before I get into the links you should know the following:
ChatML format seems to be the best fit here.
I disagree with Sao's sampler settings because the provided settings seem to have the settle into a kind of word-salad-y prose after a few messages. Try instead one of the following settings, or drop your Min P LOW. Like 0.02.
Min P: 0.02
Rep Penalty: 1.08
Presence Penalty: 1.1-1.11
OR
Min P: 0.02
XTC: 0.15, 0.3
DRY: 0.3, 1.75, 2
Raise the Min P slowly until you hit your sweet spot. Good luck, Tarnished.
This is the EXL2 4bpw version of this model. You can find the original model here.
Elder Sister Versions for Heavier Use!
14B-Qwen2.5-Kunou-v1
I do not really have anything planned for this model other than it being a generalist, and Roleplay Model? It was just something made and planned in minutes.
This is the little sister variant, the small 14B version.
Kunou's the name of an OC I worked on for a couple of years, for a... fanfic. mmm...
A kind-of successor to my smaller model series. It works pretty nicely I think?
This version is basically a better, more cleaned up Dataset used on Euryale and Stheno.
Recommended Model Settings | Look, I just use these, they work fine enough. I don't even know how DRY or other meme samplers work. Your system prompt matters more anyway.
Prompt Format: ChatML
Temperature: 1.1
min_p: 0.1
Special thanks to my wallet for funding this, my juniors who share a single braincell between them, and my current national service.
There have been more stabbings, accidents out there this winter season. It's wild. Stay safe everyone.
Also sorry for the inactivity. Life was in the way. It still is, just less so, for now. Burnout is a thing, huh?
https://sao10k.carrd.co/ for contact.
View 72B version for Axolotl Configs used.
- Downloads last month
- 19