My first second merge and model ever! Literally depraved.
I made the configuration simpler and replaced the personality model. I think it fits nicely now and produce an actual working model.
But since this is my first model, I kept thinking that it's very good.
Don't get me wrong, it is "working as intended" but yeah, try it out for yourself!
It is very very VERY obedient, it will do what you tell it to do! Works both on assistant mode and roleplay mode.
It has a deep understanding of personality and quite visceral on describing organs, the narrative feel wholesome but explicit when you need it.
I personally like it very much and fitting my use case.
I would personally thank you sleepdeprived3 for the amazing finetunes,
DoppelReflEx for giving me the dream to make a merge model someday,
and the people in BeaverAI Club for the great inspirations.
Luv <3
Addendum
I tested this model both personally and in UGI Leaderboard.
The reason why the model doesn't follow your system prompt is because the model itself is not following order (see direct willingness).
This theory is also being proved by my testing of using Venice.ai chat and seeing that the model are able to output their system prompt
when the content itself say otherwise. It is very bad at following order but perfect at adhering.
After blindly merging, I have a clear goal of what I want to make of the next generations of the model.
A good storyteller that understand personality and organs.
I am doing internal research right now, the most promising base is
TroyDoesAI/BlackSheep-24B as of writing this.
Thank you again, for everyone that helped!
DontPlanToEnd for the test and leaderboard.
mradermacher for the GGUF-s.
Artus for being a daddy (?).
DarkHazard for the BRAND NEW MODEL? (Inspired by this repo... but I'm just a newbie)
I personally use the nsigma 1.5 and temp 4 with the rest neutralized.
A bit silly yeah, but add just a tiny bit of min_p (if you want) and then turn up the XTC and DRY.
